/cdn.vox-cdn.com/uploads/chorus_image/image/46990094/usa-today-8239285.0.jpg)
This year's 128-team season preview series is complete! (I even tacked on a 128-team POWER RANKINGS piece at the end, which I'm sure people will react to with calmness, tact, and measured takes.)
Now that I've got a little bit of time on my hands, I thought it would be interesting to explore one of the aspects of the previews that I found the most interesting: percentiles. In each preview, I presented something like this:
2014 Schedule & Results
Record: 6-7 | Adj. Record: 9-4 | Final F/+ Rk: 31 | |||||||
Date | Opponent | Opp. F/+ Rk | Score | W-L | Percentile Performance |
Adj. Scoring Margin |
Win Expectancy |
1-Sep | at Louisville | 23 | 13-31 | L | 47% | -1.6 | 13% |
6-Sep | Florida A&M | N/A | 41-7 | W | 91% | 30.8 | 100% |
13-Sep | Arkansas State | 66 | 41-20 | W | 97% | 43.8 | 100% |
20-Sep | at Nebraska | 30 | 31-41 | L | 62% | 7.2 | 35% |
27-Sep | Duke | 51 | 22-10 | W | 94% | 36.3 | 99% |
4-Oct | Georgia Tech | 8 | 17-28 | L | 54% | 2.3 | 23% |
11-Oct | Cincinnati | 47 | 55-34 | W | 99% | 52.4 | 100% |
23-Oct | at Virginia Tech | 33 | 30-6 | W | 99% | 52.3 | 100% |
1-Nov | North Carolina | 70 | 47-20 | W | 98% | 48.0 | 100% |
15-Nov | Florida State | 15 | 26-30 | L | 56% | 3.7 | 32% |
22-Nov | at Virginia | 39 | 13-30 | L | 10% | -30.0 | 0% |
29-Nov | Pittsburgh | 43 | 23-35 | L | 22% | -18.1 | 1% |
27-Dec | vs. South Carolina | 38 | 21-24 | L | 30% | -12.0 | 10% |
This was similar to the Adj. Score measure that I shared in previous years' previews, but I think it communicates a clearer message. The idea was to take the key, Five Factors-related stats from each game and, instead of presenting it as a "single-game S&P+" type of score, present it in terms of national percentiles. It's a concept most of us are familiar with because of standardized tests, and it communicates a complex set of stats in a simple, intuitive manner. I like it.
Since the percentiles ARE basically a single-game S&P+ score, there's a very strong correlation between a team's average percentile performance and its full-season S&P+ rating. But I thought it would be interesting to break these percentiles out a bit further.
Below, you'll find teams' maximum, minimum, median, and mean percentile scores, along with each team's week-to-week variance. These numbers will give you a pretty good explanation of how each team finished where it finished.
Here are the correlations between each of these numbers and a team's full-season S&P+ rating:
Max | 0.774 |
Min | 0.774 |
Median | 0.972 |
Mean | 0.993 |
Variance | -0.008 |
As you would expect, the mean matches up nearly perfectly with the full-season number. The only reason the correlation isn't a perfect 1.000 is that each game has a different number of plays, and the full-season number is still based in part on play-by-play data.
But looking at max, min, and median tells us some interesting things about certain teams (like, say, Ole Miss). Here's the complete team list, sorted by S&P+ rank:
Team | Max | Rk | Min | Rk | Median | Rk | Mean | Rk | Variance | Rk | Overall S&P+ | Rk |
Ohio State | 100% | 1 | 49% | 3 | 95% | 1 | 90% | 2 | 2% | 3 | 30.2 | 1 |
Alabama | 100% | 2 | 51% | 1 | 94% | 5 | 90% | 1 | 2% | 1 | 28.3 | 2 |
Oregon | 99% | 23 | 46% | 5 | 95% | 3 | 85% | 3 | 3% | 17 | 23.7 | 3 |
Auburn | 99% | 11 | 46% | 4 | 88% | 8 | 85% | 4 | 3% | 9 | 23.6 | 4 |
Arkansas | 100% | 3 | 37% | 11 | 88% | 9 | 80% | 8 | 5% | 37 | 23.1 | 5 |
Ole Miss | 100% | 6 | 11% | 50 | 95% | 2 | 79% | 9 | 9% | 114 | 23.0 | 6 |
Georgia | 99% | 14 | 29% | 16 | 94% | 4 | 82% | 5 | 5% | 41 | 22.6 | 7 |
TCU | 100% | 5 | 34% | 12 | 89% | 6 | 80% | 6 | 3% | 18 | 19.4 | 8 |
UCLA | 99% | 21 | 42% | 7 | 81% | 16 | 78% | 11 | 4% | 21 | 19.0 | 9 |
Georgia Tech | 97% | 36 | 49% | 2 | 86% | 12 | 80% | 7 | 3% | 13 | 18.7 | 10 |
Michigan State | 99% | 20 | 45% | 6 | 87% | 10 | 79% | 10 | 4% | 26 | 18.6 | 11 |
Mississippi State | 99% | 24 | 29% | 15 | 83% | 14 | 78% | 13 | 5% | 43 | 17.8 | 12 |
Oklahoma | 100% | 7 | 23% | 22 | 82% | 15 | 75% | 15 | 6% | 66 | 17.7 | 13 |
Baylor | 98% | 31 | 30% | 14 | 86% | 11 | 78% | 12 | 5% | 32 | 17.2 | 14 |
Stanford | 98% | 28 | 28% | 17 | 88% | 7 | 75% | 17 | 7% | 81 | 16.6 | 15 |
LSU | 99% | 10 | 14% | 41 | 84% | 13 | 73% | 20 | 9% | 109 | 16.5 | 16 |
Clemson | 100% | 4 | 27% | 18 | 70% | 33 | 72% | 21 | 6% | 64 | 15.5 | 17 |
Boise State | 99% | 19 | 18% | 31 | 75% | 28 | 74% | 18 | 6% | 62 | 15.4 | 18 |
Tennessee | 99% | 12 | 18% | 32 | 70% | 34 | 69% | 28 | 7% | 87 | 14.2 | 19 |
USC | 100% | 9 | 21% | 27 | 77% | 23 | 71% | 23 | 7% | 82 | 14.2 | 20 |
Marshall | 95% | 47 | 39% | 10 | 78% | 20 | 75% | 16 | 3% | 16 | 14.1 | 21 |
Florida State | 95% | 48 | 40% | 9 | 80% | 17 | 76% | 14 | 2% | 5 | 13.7 | 22 |
Missouri | 99% | 26 | 20% | 29 | 77% | 21 | 71% | 22 | 6% | 58 | 13.0 | 23 |
Louisville | 96% | 45 | 42% | 8 | 76% | 24 | 73% | 19 | 3% | 14 | 12.9 | 24 |
Wisconsin | 98% | 29 | 2% | 113 | 77% | 22 | 70% | 24 | 8% | 98 | 12.8 | 25 |
Team | Max | Rk | Min | Rk | Median | Rk | Mean | Rk | Variance | Rk | Overall S&P+ | Rk |
Miami-FL | 99% | 18 | 10% | 52 | 62% | 46 | 66% | 32 | 10% | 121 | 12.7 | 26 |
Notre Dame | 98% | 30 | 5% | 75 | 75% | 27 | 70% | 25 | 7% | 86 | 12.1 | 27 |
Kansas State | 98% | 35 | 22% | 23 | 73% | 31 | 69% | 27 | 6% | 49 | 11.9 | 28 |
Florida | 96% | 43 | 21% | 26 | 69% | 35 | 69% | 26 | 6% | 59 | 11.6 | 29 |
Texas A&M | 98% | 32 | 4% | 86 | 60% | 51 | 66% | 35 | 9% | 111 | 10.4 | 30 |
Nebraska | 94% | 54 | 9% | 57 | 79% | 18 | 68% | 30 | 8% | 94 | 10.3 | 31 |
Arizona State | 94% | 56 | 13% | 45 | 79% | 19 | 68% | 29 | 8% | 95 | 10.2 | 32 |
Texas | 99% | 15 | 14% | 40 | 75% | 26 | 65% | 37 | 8% | 107 | 10.0 | 33 |
Louisiana Tech | 99% | 13 | 5% | 81 | 68% | 36 | 64% | 40 | 10% | 120 | 9.9 | 34 |
Arizona | 96% | 46 | 16% | 37 | 64% | 39 | 66% | 33 | 7% | 71 | 9.5 | 35 |
Minnesota | 99% | 17 | 21% | 25 | 59% | 52 | 64% | 38 | 6% | 54 | 9.1 | 36 |
Virginia Tech | 93% | 58 | 8% | 59 | 75% | 25 | 66% | 31 | 7% | 74 | 8.9 | 37 |
Cincinnati | 96% | 42 | 7% | 68 | 73% | 30 | 66% | 34 | 7% | 83 | 8.8 | 38 |
Boston College | 90% | 77 | 25% | 20 | 65% | 38 | 66% | 36 | 4% | 27 | 8.4 | 39 |
West Virginia | 99% | 25 | 31% | 13 | 62% | 48 | 63% | 41 | 4% | 22 | 8.0 | 40 |
Utah | 98% | 33 | 15% | 39 | 64% | 40 | 63% | 42 | 7% | 88 | 8.0 | 41 |
Virginia | 98% | 34 | 26% | 19 | 62% | 49 | 63% | 43 | 6% | 55 | 7.9 | 42 |
South Carolina | 95% | 52 | 23% | 21 | 73% | 32 | 64% | 39 | 6% | 50 | 7.9 | 43 |
Pittsburgh | 99% | 16 | 17% | 34 | 54% | 60 | 62% | 44 | 7% | 77 | 7.9 | 44 |
Michigan | 100% | 8 | 9% | 54 | 58% | 55 | 58% | 50 | 10% | 116 | 6.8 | 45 |
Penn State | 97% | 39 | 7% | 62 | 59% | 53 | 60% | 47 | 7% | 76 | 6.1 | 46 |
BYU | 95% | 50 | 13% | 43 | 62% | 45 | 61% | 46 | 7% | 72 | 6.1 | 47 |
Navy | 99% | 22 | 19% | 30 | 54% | 62 | 58% | 48 | 7% | 78 | 5.7 | 48 |
Memphis | 91% | 69 | 4% | 88 | 66% | 37 | 61% | 45 | 6% | 60 | 5.1 | 49 |
NC State | 93% | 57 | 1% | 114 | 74% | 29 | 58% | 49 | 11% | 124 | 4.0 | 50 |
Team | Max | Rk | Min | Rk | Median | Rk | Mean | Rk | Variance | Rk | Overall S&P+ | Rk |
Western Kentucky | 97% | 41 | 7% | 69 | 62% | 47 | 57% | 52 | 7% | 84 | 3.9 | 51 |
Utah State | 94% | 53 | 4% | 87 | 56% | 57 | 58% | 51 | 8% | 89 | 3.6 | 52 |
Maryland | 93% | 61 | 8% | 60 | 63% | 41 | 57% | 53 | 8% | 91 | 3.3 | 53 |
Air Force | 82% | 98 | 8% | 61 | 63% | 43 | 55% | 55 | 6% | 57 | 2.2 | 54 |
North Carolina | 97% | 37 | 7% | 67 | 57% | 56 | 54% | 57 | 8% | 104 | 2.1 | 55 |
Colorado State | 91% | 74 | 4% | 90 | 59% | 54 | 55% | 56 | 8% | 103 | 2.0 | 56 |
Georgia Southern | 85% | 90 | 2% | 108 | 61% | 50 | 55% | 54 | 7% | 79 | 2.0 | 57 |
Kentucky | 93% | 62 | 5% | 84 | 56% | 58 | 53% | 58 | 11% | 127 | 1.5 | 58 |
Arkansas State | 96% | 44 | 8% | 58 | 50% | 73 | 51% | 60 | 8% | 102 | 1.0 | 59 |
UL-Lafayette | 89% | 80 | 5% | 79 | 63% | 44 | 52% | 59 | 9% | 108 | 0.4 | 60 |
California | 93% | 60 | 21% | 24 | 50% | 72 | 50% | 69 | 4% | 25 | 0.3 | 61 |
Western Michigan | 92% | 66 | 7% | 70 | 48% | 75 | 50% | 64 | 8% | 100 | 0.1 | 62 |
Oklahoma State | 91% | 70 | 7% | 66 | 51% | 69 | 50% | 62 | 8% | 101 | 0.0 | 63 |
Texas Tech | 85% | 93 | 20% | 28 | 48% | 74 | 50% | 66 | 4% | 23 | 0.0 | 64 |
Illinois | 91% | 71 | 13% | 47 | 52% | 67 | 50% | 70 | 8% | 99 | -0.1 | 65 |
Temple | 92% | 65 | 7% | 63 | 52% | 68 | 50% | 71 | 8% | 105 | -0.3 | 66 |
Toledo | 86% | 89 | 13% | 46 | 53% | 64 | 50% | 63 | 6% | 52 | -0.3 | 67 |
Washington State | 87% | 86 | 14% | 42 | 47% | 77 | 49% | 73 | 6% | 70 | -0.5 | 68 |
Rutgers | 92% | 63 | 6% | 72 | 50% | 71 | 50% | 65 | 10% | 119 | -0.6 | 69 |
Iowa | 95% | 49 | 2% | 105 | 54% | 63 | 50% | 61 | 10% | 118 | -0.6 | 70 |
Central Florida | 91% | 73 | 4% | 93 | 55% | 59 | 50% | 68 | 8% | 97 | -0.8 | 71 |
East Carolina | 87% | 85 | 17% | 33 | 53% | 65 | 48% | 74 | 5% | 29 | -0.9 | 72 |
Duke | 91% | 72 | 3% | 101 | 54% | 61 | 50% | 67 | 6% | 63 | -1.0 | 73 |
Oregon State | 93% | 59 | 13% | 44 | 45% | 80 | 47% | 76 | 8% | 93 | -1.1 | 74 |
Washington | 85% | 91 | 10% | 51 | 51% | 70 | 48% | 75 | 8% | 92 | -1.3 | 75 |
Team | Max | Rk | Min | Rk | Median | Rk | Mean | Rk | Variance | Rk | Overall S&P+ | Rk |
Rice | 90% | 75 | 4% | 91 | 63% | 42 | 49% | 72 | 11% | 126 | -1.5 | 76 |
Colorado | 85% | 92 | 10% | 53 | 45% | 79 | 47% | 77 | 5% | 30 | -1.9 | 77 |
Syracuse | 98% | 27 | 16% | 36 | 37% | 90 | 43% | 87 | 5% | 45 | -2.0 | 78 |
Purdue | 87% | 82 | 15% | 38 | 47% | 78 | 46% | 78 | 6% | 61 | -2.1 | 79 |
Northern Illinois | 90% | 76 | 12% | 48 | 42% | 84 | 45% | 81 | 5% | 36 | -2.2 | 80 |
Nevada | 78% | 108 | 16% | 35 | 42% | 85 | 45% | 82 | 4% | 19 | -2.2 | 81 |
South Alabama | 97% | 40 | 3% | 99 | 43% | 83 | 46% | 79 | 10% | 122 | -2.8 | 82 |
San Diego State | 87% | 84 | 5% | 83 | 47% | 76 | 45% | 80 | 6% | 51 | -2.9 | 83 |
UAB | 88% | 81 | 7% | 64 | 35% | 93 | 44% | 84 | 9% | 112 | -3.0 | 84 |
Northwestern | 94% | 55 | 3% | 97 | 44% | 82 | 43% | 85 | 8% | 106 | -3.9 | 85 |
Indiana | 92% | 67 | 6% | 74 | 52% | 66 | 45% | 83 | 10% | 117 | -3.9 | 86 |
Middle Tennessee | 82% | 97 | 11% | 49 | 40% | 88 | 42% | 90 | 6% | 53 | -4.3 | 87 |
UTEP | 97% | 38 | 1% | 116 | 41% | 87 | 42% | 88 | 11% | 125 | -4.5 | 88 |
Central Michigan | 84% | 94 | 1% | 123 | 36% | 92 | 43% | 86 | 7% | 85 | -5.4 | 89 |
Houston | 81% | 102 | 1% | 122 | 41% | 86 | 42% | 89 | 6% | 56 | -5.8 | 90 |
New Mexico | 79% | 103 | 9% | 55 | 31% | 97 | 38% | 92 | 5% | 46 | -6.5 | 91 |
Kansas | 92% | 64 | 4% | 95 | 30% | 100 | 38% | 91 | 9% | 113 | -7.1 | 92 |
Tulane | 84% | 95 | 7% | 65 | 29% | 103 | 36% | 94 | 6% | 68 | -7.5 | 93 |
UL-Monroe | 79% | 104 | 1% | 115 | 44% | 81 | 37% | 93 | 5% | 28 | -8.0 | 94 |
Iowa State | 78% | 106 | 5% | 76 | 37% | 91 | 35% | 96 | 5% | 40 | -8.3 | 95 |
Bowling Green | 86% | 87 | 4% | 92 | 31% | 98 | 34% | 100 | 6% | 67 | -9.1 | 96 |
Florida Atlantic | 87% | 83 | 3% | 98 | 32% | 96 | 35% | 97 | 7% | 75 | -9.1 | 97 |
Ohio | 92% | 68 | 4% | 94 | 29% | 102 | 34% | 99 | 8% | 96 | -9.4 | 98 |
Miami-OH | 68% | 118 | 5% | 77 | 34% | 94 | 32% | 104 | 3% | 12 | -9.6 | 99 |
Fresno State | 73% | 111 | 3% | 104 | 39% | 89 | 36% | 95 | 8% | 90 | -9.7 | 100 |
Team | Max | Rk | Min | Rk | Median | Rk | Mean | Rk | Variance | Rk | Overall S&P+ | Rk |
Kent State | 78% | 105 | 2% | 112 | 32% | 95 | 34% | 101 | 5% | 44 | -9.7 | 101 |
Old Dominion | 66% | 120 | 5% | 80 | 31% | 99 | 31% | 107 | 4% | 20 | -10.6 | 102 |
Southern Miss | 71% | 115 | 5% | 78 | 28% | 105 | 31% | 106 | 5% | 34 | -10.7 | 103 |
Akron | 86% | 88 | 2% | 109 | 16% | 119 | 33% | 102 | 9% | 115 | -10.9 | 104 |
Vanderbilt | 73% | 110 | 3% | 103 | 28% | 104 | 31% | 105 | 5% | 38 | -10.9 | 105 |
Texas State | 82% | 99 | 5% | 85 | 22% | 112 | 30% | 108 | 6% | 69 | -11.0 | 106 |
Wyoming | 89% | 79 | 9% | 56 | 18% | 118 | 29% | 111 | 7% | 73 | -11.2 | 107 |
Georgia State | 77% | 109 | 3% | 100 | 27% | 106 | 30% | 109 | 5% | 47 | -11.5 | 108 |
Idaho | 67% | 119 | 6% | 71 | 27% | 107 | 27% | 114 | 2% | 7 | -11.6 | 109 |
UTSA | 95% | 51 | 1% | 126 | 30% | 101 | 33% | 103 | 9% | 110 | -12.1 | 110 |
Florida International | 71% | 114 | 2% | 110 | 19% | 116 | 30% | 110 | 6% | 65 | -12.3 | 111 |
Wake Forest | 62% | 122 | 6% | 73 | 26% | 108 | 28% | 113 | 3% | 15 | -12.4 | 112 |
Appalachian State | 83% | 96 | 0% | 128 | 25% | 110 | 35% | 98 | 11% | 123 | -12.5 | 113 |
San Jose State | 72% | 113 | 3% | 102 | 26% | 109 | 28% | 112 | 5% | 48 | -13.6 | 114 |
South Florida | 54% | 124 | 3% | 96 | 22% | 111 | 25% | 117 | 3% | 11 | -13.7 | 115 |
Ball State | 72% | 112 | 4% | 89 | 18% | 117 | 26% | 116 | 5% | 35 | -14.2 | 116 |
UNLV | 45% | 126 | 5% | 82 | 20% | 114 | 22% | 120 | 2% | 2 | -15.0 | 117 |
Tulsa | 45% | 127 | 2% | 107 | 22% | 113 | 23% | 119 | 2% | 6 | -15.5 | 118 |
Connecticut | 78% | 107 | 2% | 111 | 15% | 120 | 22% | 121 | 5% | 33 | -16.3 | 119 |
Massachusetts | 68% | 117 | 1% | 121 | 14% | 121 | 22% | 123 | 4% | 24 | -17.5 | 120 |
Troy | 70% | 116 | 1% | 117 | 14% | 122 | 22% | 122 | 5% | 42 | -18.1 | 121 |
Buffalo | 81% | 101 | 1% | 125 | 19% | 115 | 25% | 118 | 7% | 80 | -18.8 | 122 |
Hawaii | 52% | 125 | 2% | 106 | 13% | 123 | 18% | 126 | 2% | 8 | -18.8 | 123 |
SMU | 55% | 123 | 1% | 119 | 13% | 124 | 18% | 125 | 3% | 10 | -19.3 | 124 |
Army | 64% | 121 | 1% | 118 | 6% | 126 | 20% | 124 | 5% | 31 | -19.8 | 125 |
North Texas | 90% | 78 | 0% | 127 | 6% | 127 | 26% | 115 | 12% | 128 | -20.5 | 126 |
New Mexico State | 44% | 128 | 1% | 124 | 10% | 125 | 15% | 127 | 2% | 4 | -21.4 | 127 |
Eastern Michigan | 81% | 100 | 1% | 120 | 5% | 128 | 14% | 128 | 5% | 39 | -24.6 | 128 |
52 teams hit the 95th percentile at least once, which is more than I would have expected. Even more interesting and/or maddening: of these 52 teams, 15 also played at least one game at the 10th percentile or lower: Arkansas State, Cincinnati, Iowa, Louisiana Tech, Miami, Michigan, North Carolina, Notre Dame, Penn State, South Alabama, Texas A&M, UTEP, UTSA, Western Kentucky, and Wisconsin.
Meanwhile, while most of the teams at the top and bottom of the rankings had pretty low variance overall (which is why they ended up on one end or the other on the bell curve), a few bucked that trend. At the high end, Ole Miss, Oklahoma, Stanford, and LSU had pretty crazy levels of variance, as did Tennessee and USC. Meanwhile, at the lower end, teams like Appalachian State (which improved dramatically at the end of the year), Buffalo, and North Texas experienced pretty high highs to go with steady lows.
What does this mean for 2015? Well ... I'm not sure. This is a pretty new concept, but it will be interesting to keep working with this to see if the volatile teams are easier or (more likely) harder to project the next season, and if something like median is more effective in projections than mean.
UPDATE: Two more points I originally intended to make.
1. As mentioned at Tomahawk Nation, consistency saved Florida State. The Seminoles never really put together a perfect performance and were stuck in the teens in terms of both median and mean, but their extreme consistency bailed them out. They were in the 80th to 85th percentile seemingly every single week and didn't lay an outright egg until the Rose Bowl.
2. The fact that 78 different teams hit the 90th percentile at least once (including North Texas, UTSA, and, yes, Kansas), and every team hit the 51st percentile or lower at least once, reminds us why head-to-head results are a bit overrated. College football teams are bipolar, and just because you were better than a team one week doesn't mean you would be the other 13 or so. Head-to-head is a good conference tie-breaker, but that's about it.
Loading comments...