Rank Team Win Pct. Sagarin
1 Michigan 0.73414 11
2 Notre Dame (IN) 0.73096 26
3 Boise St. (ID) 0.72190 9
4 Oklahoma 0.71531 4
5 Texas 0.71429 17
6 Ohio St. 0.71315 49
7 Alabama 0.70986 1
8 Nebraska 0.70345 25
9 Southern California 0.70332 12
10 Penn St. 0.68953 28
11 Tennessee 0.68703 50
12 Florida St. 0.66414 23
13 Louisiana St. 0.64688 2
14 Georgia 0.64476 21
15 Auburn (AL) 0.63376 33
16 Miami (FL) 0.63172 52
17 Florida 0.62980 29
18 Miami (OH) 0.61559 98
19 Washington 0.61194 44
20 South Florida 0.61017 59
21 Virginia Tech 0.60940 27
22 Arizona St. 0.60487 41
23 West Virginia 0.60141 22
24 Central Michigan 0.60091 137
25 Texas A&M 0.59831 14
26 Southern Mississippi 0.59740 24
27 Colorado 0.59622 107
28 Georgia Tech 0.59556 56
29 Arkansas 0.59489 6
30 Utah 0.59352 39
As you see, such a list leaves much to be desired in the completeness category, and unless we have all 120 teams on such a list it will need some streamlining to get even today’s Top 25 (or whatever) covered.
I suppose a case could be made for using the final AP polls (and other such polls) over the past 10 years or so and building a list of how often and at what level each team made the lists.
A sample of such is at College Football Poll.com and the years of 2010 and 2011 would need to be added.
I can’t remember where, but I think I have seen such analysis before.
Here’s an idea, but I have no clue as to where to get the data to actually follow through with it.
Instead of looking at a team’s own performance, look at the performance of their opponents in the games versus that team, in comparison with their average performance in the rest of their games.
For this example I’m going to use Net Yardage (yards gained minus yards allowed). Ideally you’d do this with a bunch of different stats and see if anything “pops”.
Let’s look at the Virginia Tech at Georgia Tech game on Nov 20.
We’re ranking the Hokies. Georgia Tech’s net yardage in this game was -136. Georgia Tech’s net yardage in all their other games was +361, +254, +406, +164, +26, +53, -111, -51, +44, +198, -25, +111. So at this point there’s a few things we can do to get an idea of how much better or worse Georgia Tech played than their norm. I’ll just take the mean. Georgia Tech averages 119 net yards in all their games aside from the VT one. So Virginia Tech did 255 yards better than GT’s non-VT average.
Do this for all of Virginia Tech’s games and you can get an “average net yards versus the opponent’s non-VT mean”. Then do this for all teams, and you can rank them.
There’s no doubt that your method would improve on the bare-bones W-L stats for any team, Jilaad. And I would approve of the results of such a method over placement in some poll or group of polls if it came to a comparison of results.
I suppose my main issue is: how many years are reasonable to use in determining a team’s overall (or all-time) strength level? Given that the strength of an individual team varies from year to year due to turnover of coaches, players, conference management and other factors, how long does it take for that “All-Time Power” classification to mean something?
Compare the end-of-season Final BCS Rankings with the latest Sagarin and the All-Time Win% rankings:
Dec. 4, 2011 Final BCS standings
Rank Team Sagarin All-Time Win%
1 LSU 2 13
2 Alabama 1 7
3 Oklahoma State 3 80
4 Stanford 7 46
5 Oregon 5 57
6 Arkansas 6 29
7 Boise State 9 3
8 Kansas State 19 110
9 South Carolina 10 79
10 Wisconsin 8 42
11 Virginia Tech 27 21
12 Baylor 13 84
13 Michigan 11 1
14 Oklahoma 4 4
15 Clemson 32 32
16 Georgia 21 14
17 Michigan State 18 31
18 TCU 20 65
19 Houston 15 64
20 Nebraska 25 8
21 Southern Mississippi 24 26
22 Penn State 28 10
23 West Virginia 22 23
24 Texas 17 5
25 Auburn 33 15
It appears to me that West Virginia is the only team whose rankings are consistent enough to declare reliable. And how far back does #23 reliably state their ranking overall?
One other comment on your approach: that’s a lot of number-crunching even if you can locate the source(s) for all that data for a good number of years!
Oops, I totally forgot that you were looking for indicators of long-term strength. My method would only really be appropriate to use within a single season, since the game changes so much from year to year.
The year to year change is going to be a problem in any multi-year ranking system. I think the best way to do it would be to rank each season separately using a ranking scheme that you trust, then average each team’s ranks across whatever time period you’re looking at.
If we do that using the last 5 years of the Colley Matrix rankings, this is what we get:
rank team avg_yearly_rank
1 OREGON 10.4
2 OKLAHOMA 10.4
3 LSU 12.4
4 ALABAMA 12.8
5 VA TECH 13
6 USC 14
7 BOISE ST 14
8 SOUTHERN CAL 14.5
9 FLORIDA 16.6
10 TCU 17.8
11 OHIO STATE 19.2
12 WEST VIRGINIA 19.8
13 OKLAHOMA ST 20.6
14 MISSOURI 20.8
15 TEXAS 21.8
16 UTAH 22.4
17 PENN STATE 23.6
18 GEORGIA 24.6
19 WISCONSIN 26
20 BRIGHAM YOUNG 26.2
21 ARKANSAS 27.2
22 NEBRASKA 28.8
23 CINCY 29
24 AUBURN 29.6
25 S CAROLINA 30.8
So the list seems okay. Good teams are at the top, worse teams are at the bottom. But there’s some weirdness. I don’t think anybody would agree that Alabama has been just a teeny bit better than Virginia Tech over the 5 year period - they’ve been overwhelmingly better. However, when you look deeper into it, Alabama has been super dominant the last 4 years (ranked 3, 13, 1, and 6) but in 2007 they were ranked 41. Virginia Tech ranks consistently in the teens (16, 16, 10, 15, 8).
That one crap year for Alabama is killing them in the 5 year ranking. This suggests that taking a straight average across the 5 years might not be a good way to do it. Maybe do something like figure skating judging, where you toss the best and worst. Or maybe do some kind of a weighted average to increase the importance of top 10 or top 5 finishes.
Or maybe the Colley Matrix is just a bad ranking system to use. I picked it for the example because the data is readily available and includes all teams.
Thanks for the mention of the Colley Matrix. That’s the sort of thing I’m looking for and the pre-crunched numbers aspect is a big selling point. I enjoy toying with the numbers but if somebody else has already done so, I’m good with that.
Are there other such systems that have some measure of credibility and respect? I would really like to find a more reliable system than Sagarin’s!
Here is the home page for the Billingsley Rankings, and even though it looks like a late 90’s Geocities site it is actually one of the computer rankings used by the BCS. The page also has links to all of the other BCS computers rankings.
You might be interested in Football Outsiders. Their focus is on the NFL game, although they do have a weekly college column. They primarily use a statistic based on per play success rate (A first down play is a success if it get 45% of the needed yards to gain a first down/TD, a 2nd down play is a success if it gets 60%, 3rd and 4th down are a success only if you actually convert). I linked to the page that explains their stats.
Thanks again, Jilaad. After some further searching I found AP Season Appearance Streaks: Any / Active only which is similar to (if not the same as) what I was thinking of about the AP Poll appearances as a measure of long-term strength.
It may mean something (not sure just what) as an additional column in something like a combination of the charts we’ve already built and posted. Some “weighting” technique will need to be used to make recent power outweigh long-term or years-ago strength. Or does that really apply when you think of how Auburn (as an example) went from Undefeated National Champs to an 8-5 team the very next year?
In short, what criteria need to be considered when we treat the idea of “team” as if it had meaning beyond the scope of one to four years, when roughly the same players and coaches had anything to do with the success of the program?
I found a method that might actually work! So the concept is to assign “dominance points”, so a #1 or #2 finish will be worth more dominance points than a #10 or #15 performance, and will be worth enough to diminish the occasional bad year that all teams have. This is to eliminate the Alabama/VT problem from my first attempt at a long term rank.
My first attempt to assign dominance points (and my only one, because this ended up working FANTASTIC) was DominancePoints = 1/Rank. So an end of year rank of 1 gives you one dominance point. A rank of 2 gives you only 0.5 dominance points - it takes two years at number 2 to count as being as dominant as one year at number one. I’m sure the formula can be tweaked, but we might not need to. Here are the top 25 results.
rank team total
1 LSU 1.759649123
2 FLORIDA 1.649482945
3 ALABAMA 1.601313321
4 OKLAHOMA ST 1.218304787
5 AUBURN 1.141195073
6 TEXAS 0.956439394
7 OKLAHOMA 0.916666667
8 TCU 0.835267212
9 GEORGIA 0.706654457
10 OREGON 0.56031746
11 BOISE ST 0.554047229
12 OHIO STATE 0.527777778
13 MISSOURI 0.522034813
14 SOUTHERN CAL 0.503781004
15 UTAH 0.50085603
16 CINCY 0.425137363
17 WEST VIRGINIA 0.420736671
18 VA TECH 0.416666667
19 STANFORD 0.383240391
20 ARKANSAS 0.371432063
21 PENN STATE 0.301526252
22 S CAROLINA 0.285212755
23 KANSAS 0.264850618
24 TEXAS TECH 0.254402709
25 BRIGHAM YOUNG 0.253465429
I used the Colley Matrix over the last 5 years again, because I still had the data from the first attempt.
The goals I was looking for in a ranking that makes sense were: Alamaba and Florida should be at or near the top, Virginia Tech could be anywhere in the teens but well ahead of any other ACC team (the next ACC team was Boston College at 32), Boise State should be ahead of VT but not overwhelmingly so, Oregon should be high but not top 5 (the “average of ranks” from before had them in a tie for #1 which was fishy). All goals were met. I’m excited about this.
That is genuinely impressive! It would appear that additional tweaking may be called for, especially by the B1G aficionados, but as a first run, and from only one data source, that does have some good-feeling results.
I just hope some of those passing through will offer some comments as to where and how to tweak your idea. I like it.