Writers of Pro Football Prospectus 2008

Most Recent FO Features

JohnsonKer18.jpg

» Seventh Day Adventure: Week 13

The biggest game this week is the Iron Bowl, where the playoff hopes of Alabama, Auburn, and Georgia hang in the balance.

07 Nov 2014

VN: A Big List of Factors

by Bill Connelly

Last January, I posted what became one of my more well-read pieces at Football Study Hall: the Five Factors.

[O]ver time, I've come to realize that the sport comes down to five basic things, four of which you can mostly control. You make more big plays than your opponent, you stay on schedule, you tilt the field, you finish drives, and you fall on the ball. Explosiveness, efficiency, field position, finishing drives, and turnovers are the five factors to winning football games.

* If you win the explosiveness battle (using PPP), you win 86 percent of the time.
* If you win the efficency battle (using Success Rate), you win 83 percent of the time.
* If you win the drive-finishing battle (using points per trip inside the 40), you win 75 percent of the time.
* If you win the field position battle (using average starting field position), you win 72 percent of the time.
* If you win the turnover battle (using turnover margin), you win 73 percent of the time.

In the following months, I tried to unpack as many of these five factors as possible. I stripped efficiency from explosiveness, creating ways to look at each that don't double-count the efficiency component. I looked at the most direct components of field position. (Spoiler alert: they're exactly what you think they would be.) I looked at the most effective way to measure a team's drive-finishing ability.

I didn't get nearly as far as I wanted before the season began, but as I worked through this, I always had a potential S&P+ (and, therefore, F/+) redesign in the back of my head. Unpack each of these five factors, create a list of basically 10 to 15 subfactors, then play with ways to measure them within a new S&P+ structure.

That's certainly going to be the goal for the coming offseason, but I wanted to take a sneak peek at where we might be headed with this. So it's time for a bit of an experiment.

Below is a table with quite a few measures, all either unadjusted for opponent or culled from other sources, i.e., Brian Fremeau's special teams ratings. (I have my own ideas for determining special teams efficiency, explored a bit here, but I haven't fleshed them out yet, and besides, Brian's numbers are good.) Here's a list, along with the "Factors" associated with them.

  • Success Rate Margin (offensive success rate minus defensive success rate) -- efficiency, field position, finishing drives, perhaps turnovers (because turnovers are more likely to come from passing downs, and efficiency helps you to avoid passing downs).

  • IsoPPP Margin -- explosiveness, field position. This is described here. Essentially, it is a measure of the magnitude of only a team's successful plays. This isolates efficiency from explosiveness by looking only at what happens once you've been efficient.
  • Red zone Success Rate Margin -- finishing drives. In its current state, this causes a bit of overlap, since the overall success rate figures above also include red zone plays. Once the offseason comes, I will look at isolating the two, but we'll go with this for now.
  • Five FEI special teams stats (Kick Efficiency, Kick Return Efficiency, Punt Efficiency, Punt Return Efficiency, Field Goal Efficiency) -- field position (for the first four), finishing drives (for FGE). Special teams play a small but undeniable role in a game's outcome, and aside from how it impacts the drive component of S&P+, it doesn't play much of a role in S&P+. It will in the future iteration.
  • Sack Rate Margin -- turnovers. So here's where things get messy. Turnovers are a mix of skill, randomness, and luck, and unpacking the former one from the latter two is quite hard. I have played with INT-to-PBU rates, standard downs-to-passing downs ratios, etc, but of the primary stats at our disposal, the only thing directly connected to turnovers is sacks. Sacks lead to forced fumbles at a higher rate than the normal play, plus I figure there's the indirect effect of telling you how harried a passer is from how often he's getting brought to the ground. A harried passer is a more pick-prone passer, right? I didn't expect this to play a role, but it might play a pretty strong one.
  • Expected Turnover Margin and Turnovers Luck -- turnovers. This looks at the number of fumbles, interceptions, and passes broken up in a given game (for both offenses) and uses national averages to assign probabilities. Normally, interceptions make up between 21 and 23 percent of total passes defensed at the college level. (So far in 2014, it's around 21 percent.) Meanwhile, offenses tend to recover about 51 to 53 percent of fumbles, though there's a way to dial in a bit more specifically there. Based on general averages, you can determine about how many turnovers a team should have had, then compare it to how many they did have, which is where Turnovers Luck comes into play. Turnovers Luck, for these purposes, is basically the difference between projected and actual TO margin.

Using these factors, I wanted to see how close I could get to matching a team's percentage of points scored for the season. Using a simple S&P calculation of 80 percent success rate and 20 percent IsoPPP gives you a correlation of 0.896 to a team's percentage of points, or about 80.3 percent of the variance. That's quite good. But can we get closer by factoring in special teams, red zone execution, and those messy turnovers? Yep!

Now, this risks some massive overfitting -- begs for it, actually -- and I will deal with that with a larger data set in the offseason. But out of pure curiosity, I wanted to overfit the hell out of this just to see how close I could get to a 1.000 correlation. Since we're dealing with different types of numbers (percentages, averages, Fremeau's numbers, etc.) I broke everything down to standard deviations above and below the average.

I got to a 0.930 correlation (87 percent) with the following combination of factors:

  • 33 percent success rate
  • 23 percent IsoPPP
  • 11 percent turnovers luck
  • 8 percent red zone efficiency
  • 6 percent sack margin
  • 6 percent kickoff efficiency
  • 5 percent field goal efficiency
  • 4 percent expected turnover margin
  • 2 percent kick return efficiency
  • 2 percent punt efficiency
  • 1 percent punt return efficiency.

(That adds up to 101 percent, but it's 100 percent with proper rounding.)

So here's a huge table with each factor listed above. For space, I combined the special teams ratings into a single number ('New STE') based on the weights above. Teams are sorted by their percentage of points scored so you can see which teams match up well with this new "5 Factors" rating (presented in terms of standard deviations above or below the average) and which don't.

Team Succ. Rt.
Margin
Rk IsoPPP
Margin
Rk RZ SR
Margin
Rk 'New'
STE
Rk Sack
Margin
Rk Exp. TO
Margin
Rk TO
Luck
Rk % of
Pts
Rk 5
Factor
Rk
Marshall 11.6% 8 0.62 1 8.5% 22 0.02 55 1.8% 38 5.1 15 -3.1 98 0.736 1 2.38 1
Wisconsin 12.7% 7 0.23 7 3.2% 53 0.00 68 5.7% 5 0.4 65 -2.4 90 0.724 2 1.46 11
Alabama 16.5% 2 0.17 11 21.0% 1 -0.03 88 4.8% 7 3.5 28 -5.5 116 0.723 3 1.78 6
Ole Miss 3.9% 48 0.30 3 0.9% 68 0.10 15 -0.9% 84 2.0 43 10.0 2 0.706 4 1.42 12
Baylor 11.4% 9 0.22 8 9.4% 16 0.04 46 3.5% 18 6.9 9 1.1 52 0.697 5 1.69 8
Ohio State 16.5% 3 0.11 32 7.2% 29 0.08 24 0.2% 66 1.4 49 6.6 12 0.697 6 1.88 3
Michigan State 23.3% 1 -0.05 89 12.2% 6 -0.01 73 7.3% 1 5.9 14 6.1 13 0.692 7 2.18 2
Georgia Southern 7.0% 27 0.28 5 9.3% 17 0.05 40 1.7% 42 2.8 35 5.2 16 0.680 8 1.56 10
TCU 9.0% 15 0.13 21 10.5% 13 0.08 22 2.5% 32 -0.2 71 15.2 1 0.680 9 1.79 5
LSU 9.6% 14 -0.04 82 16.7% 3 0.10 14 -4.2% 116 1.5 48 2.5 34 0.674 10 0.93 22
Nebraska 10.3% 10 0.15 15 5.7% 42 0.03 49 0.2% 68 -0.3 74 -0.7 72 0.673 11 1.09 18
Kansas State 4.9% 41 0.22 9 3.9% 48 0.05 37 -2.9% 112 2.3 41 3.7 23 0.673 12 0.98 19
Mississippi State 15.3% 5 0.12 25 18.4% 2 0.09 20 1.2% 51 -0.1 70 1.1 51 0.660 13 1.81 4
Oklahoma 16.1% 4 0.04 55 11.4% 11 0.10 18 4.3% 13 6.3 11 -2.3 89 0.656 14 1.62 9
Memphis 3.7% 49 0.12 27 4.3% 46 0.14 3 2.8% 28 4.7 17 0.3 61 0.652 15 0.84 26
Duke -1.0% 82 0.17 12 9.6% 15 0.15 2 3.3% 19 1.3 53 6.7 11 0.650 16 0.95 20
Oregon 8.8% 16 0.29 4 14.3% 5 0.06 32 -1.9% 99 10.3 1 1.7 44 0.647 17 1.71 7
Georgia 6.9% 28 0.14 18 6.7% 33 0.05 43 -0.4% 74 4.0 21 9.0 5 0.645 18 1.28 15
Clemson 8.3% 20 0.02 62 3.6% 50 -0.07 108 6.9% 2 1.4 52 1.6 47 0.639 19 0.74 32
Auburn 7.6% 23 0.07 47 11.3% 12 -0.01 77 1.3% 50 7.9 4 -0.9 74 0.638 20 0.87 24
Houston -0.3% 76 0.09 41 -0.7% 84 -0.01 72 -0.8% 82 4.3 19 3.7 22 0.636 21 0.20 54
Louisville 8.5% 18 0.06 48 10.5% 13 0.07 26 -0.5% 78 -3.4 108 7.4 9 0.636 22 1.14 17
Florida State 8.1% 22 0.11 28 5.5% 43 0.11 9 -0.1% 72 2.5 39 -3.5 100 0.628 23 0.90 23
Notre Dame 6.8% 30 0.12 26 6.3% 36 0.02 52 0.8% 58 -2.4 96 4.4 20 0.621 24 0.94 21
Temple -2.5% 92 0.12 24 7.6% 26 0.13 5 1.9% 37 1.6 47 5.4 15 0.615 25 0.56 39
Team Succ. Rt.
Margin
Rk IsoPPP
Margin
Rk RZ SR
Margin
Rk 'New'
STE
Rk Sack
Margin
Rk Exp. TO
Margin
Rk TO
Luck
Rk % of
Pts
Rk 5
Factor
Rk
Utah -0.2% 75 0.09 43 7.2% 30 0.07 27 2.8% 28 3.1 31 4.9 18 0.611 26 0.58 36
East Carolina 8.6% 17 0.10 39 11.7% 9 -0.15 122 -0.3% 73 -5.0 114 -1.0 77 0.609 27 0.57 37
USC 1.8% 58 0.11 31 1.8% 62 -0.01 75 -2.4% 104 9.2 3 -0.2 66 0.607 28 0.33 48
Stanford 9.6% 13 0.14 17 0.9% 67 0.01 65 2.1% 35 -8.0 125 -1.0 76 0.605 29 0.83 28
Miami 8.3% 20 0.40 2 3.2% 52 -0.01 74 1.3% 47 9.6 2 -7.6 126 0.601 30 1.36 14
Missouri 4.8% 43 0.11 34 6.2% 38 0.07 28 3.0% 26 7.2 8 -0.2 67 0.598 31 0.85 25
Louisiana Tech 7.1% 25 0.12 23 8.3% 23 0.11 10 4.3% 12 0.6 62 7.4 10 0.597 32 1.39 13
Iowa 3.6% 50 0.09 42 5.2% 44 0.03 50 3.6% 16 -0.3 73 2.3 38 0.597 33 0.64 34
Arkansas 1.7% 59 0.00 70 5.8% 40 -0.10 112 -0.7% 79 -0.9 78 -0.1 65 0.593 34 -0.12 68
Arizona State 2.9% 53 0.24 6 1.3% 65 0.00 67 0.1% 70 0.8 59 2.2 39 0.588 35 0.74 30
Colorado State 7.5% 24 0.16 14 2.7% 56 0.11 12 -3.8% 114 1.9 45 -2.9 95 0.586 36 0.82 29
Texas A&M 5.9% 34 0.03 60 7.1% 31 0.05 35 4.5% 8 -1.5 86 -6.5 120 0.585 37 0.40 45
Georgia Tech 1.0% 66 0.14 19 8.6% 21 0.00 69 -0.7% 80 7.3 7 0.7 56 0.585 38 0.51 42
West Virginia 5.0% 40 -0.05 87 2.7% 57 0.12 7 -2.1% 102 -6.7 121 -5.3 114 0.585 38 -0.02 65
Boston College 4.8% 45 0.11 30 6.6% 34 0.02 58 0.0% 71 1.8 46 -1.8 83 0.577 40 0.57 38
Western Michigan 7.0% 26 0.17 13 6.5% 35 -0.09 111 1.8% 40 5.1 16 -3.1 97 0.577 41 0.73 33
Arizona -0.4% 77 0.07 46 -4.4% 98 0.02 53 3.1% 24 2.6 37 0.4 58 0.575 42 0.08 61
Air Force 8.3% 19 -0.13 113 15.4% 4 0.09 21 2.0% 36 0.6 64 3.4 25 0.573 43 0.74 31
Utah State 5.2% 37 0.20 10 1.6% 64 -0.05 102 5.0% 6 3.7 23 8.3 8 0.568 44 1.15 16
Boise State 13.2% 6 -0.11 103 11.8% 8 -0.01 71 3.7% 15 3.1 32 -2.1 87 0.559 45 0.83 27
Arkansas State -1.3% 85 0.05 50 -3.3% 94 -0.07 107 1.4% 45 -0.3 75 2.3 36 0.559 46 -0.18 72
Minnesota 4.0% 47 0.11 36 7.5% 27 0.09 19 -4.8% 120 6.4 10 -1.4 82 0.558 47 0.59 35
Washington -1.0% 83 0.11 37 1.0% 66 0.04 48 3.0% 27 1.4 49 9.6 3 0.557 48 0.56 40
Appalachian State 3.4% 52 -0.12 107 4.5% 45 -0.33 127 0.4% 62 1.0 56 2.0 42 0.557 49 -0.55 90
UCLA 5.2% 38 0.02 66 3.0% 55 0.05 36 -5.5% 121 -1.5 88 0.5 57 0.549 50 0.27 50
Team Succ. Rt.
Margin
Rk IsoPPP
Margin
Rk RZ SR
Margin
Rk 'New'
STE
Rk Sack
Margin
Rk Exp. TO
Margin
Rk TO
Luck
Rk % of
Pts
Rk 5
Factor
Rk
Florida -1.9% 86 0.02 65 0.3% 69 0.02 57 4.3% 11 4.2 20 -4.2 107 0.548 51 -0.20 73
Kentucky 0.4% 72 0.11 33 -7.8% 114 -0.04 90 -1.3% 93 -1.2 82 9.2 4 0.547 52 0.24 52
Central Florida 0.4% 71 0.14 20 0.1% 73 -0.02 80 1.8% 39 -1.0 79 -3.0 96 0.546 53 0.10 59
Penn State 2.8% 55 -0.04 85 -3.8% 96 -0.06 104 1.3% 49 -0.2 72 -2.8 94 0.545 54 -0.28 79
Nevada -5.5% 105 0.05 52 -2.5% 89 0.05 42 1.0% 55 7.9 5 3.1 27 0.543 55 -0.17 71
Pittsburgh 9.9% 12 -0.14 114 5.7% 41 0.00 70 -2.6% 107 -3.5 109 -2.5 93 0.542 56 0.13 57
Texas State -0.7% 81 0.04 54 11.9% 7 -0.02 85 0.1% 69 2.3 41 -0.3 68 0.541 57 0.08 60
Cincinnati -2.1% 91 0.13 22 6.7% 32 0.04 47 2.7% 30 0.0 69 2.0 43 0.539 58 0.32 49
BYU 5.3% 36 -0.06 95 8.0% 25 0.07 29 -2.8% 109 -2.8 101 0.8 55 0.539 59 0.25 51
Virginia Tech 6.2% 33 -0.24 120 6.3% 37 0.06 33 5.9% 4 1.4 51 -5.4 115 0.538 60 -0.09 67
UAB 4.8% 42 -0.11 104 -4.2% 97 0.10 16 -0.8% 83 3.6 27 -7.6 125 0.533 61 -0.22 74
Central Michigan 10.1% 11 0.00 75 8.7% 20 -0.07 109 -4.7% 118 0.3 66 -4.3 108 0.531 62 0.35 47
Akron 1.2% 63 -0.02 80 1.7% 63 0.11 11 2.1% 34 0.7 61 0.3 60 0.530 63 0.16 56
Rice 2.9% 54 -0.02 79 -0.2% 77 -0.05 99 4.3% 10 6.0 13 0.0 63 0.525 64 0.13 58
Virginia 2.0% 56 -0.05 91 -3.1% 93 0.01 64 6.3% 3 3.3 29 -1.3 81 0.524 65 -0.01 64
San Diego State 6.4% 31 0.03 57 4.1% 47 0.08 25 1.0% 56 -3.0 103 -1.0 78 0.524 66 0.53 41
Maryland 1.6% 60 0.04 53 9.3% 18 0.13 6 -1.2% 91 3.3 30 -2.3 88 0.519 67 0.36 46
South Alabama 5.2% 39 -0.05 93 -0.6% 83 -0.02 82 1.3% 48 6.2 12 -3.2 99 0.519 68 0.06 62
Navy -0.1% 74 0.14 16 -1.9% 86 -0.04 98 -12.9% 128 -2.6 97 -2.4 91 0.514 69 -0.40 84
South Carolina -2.1% 90 0.02 63 -0.6% 82 0.00 66 -2.0% 101 -2.7 98 -1.3 79 0.513 70 -0.41 85
California -2.1% 89 0.10 38 -2.2% 87 -0.04 96 -2.7% 108 -3.8 111 3.8 21 0.512 71 -0.16 69
Tennessee -0.5% 79 -0.04 84 -7.5% 112 0.04 45 -0.4% 75 -1.7 90 1.7 46 0.512 72 -0.33 81
Middle Tennessee -2.1% 88 0.07 45 3.8% 49 -0.04 94 -1.5% 96 4.6 18 -3.6 102 0.511 73 -0.24 76
Toledo 1.5% 61 0.00 74 -0.2% 78 0.01 63 3.3% 23 -3.7 110 -1.3 80 0.510 74 -0.06 66
N.C. State 5.4% 35 -0.07 97 11.4% 10 0.05 39 0.8% 58 0.8 58 2.2 41 0.510 75 0.48 43
Team Succ. Rt.
Margin
Rk IsoPPP
Margin
Rk RZ SR
Margin
Rk 'New'
STE
Rk Sack
Margin
Rk Exp. TO
Margin
Rk TO
Luck
Rk % of
Pts
Rk 5
Factor
Rk
Michigan 1.2% 64 0.01 69 1.8% 61 0.02 54 1.1% 52 -8.3 126 -4.7 110 0.503 76 -0.26 78
Ball State -4.7% 103 -0.13 108 -10.1% 121 0.08 23 -0.5% 76 2.6 37 8.4 7 0.502 77 -0.51 88
UTEP 4.5% 46 -0.26 122 8.2% 24 -0.10 115 1.1% 53 7.4 6 1.6 49 0.501 78 -0.25 77
UL-Lafayette 6.8% 29 -0.10 102 6.1% 39 0.03 51 4.1% 14 -2.8 99 -4.2 106 0.497 79 0.16 55
Texas 3.6% 51 0.02 64 7.3% 28 -0.13 120 1.5% 44 -2.8 100 2.8 33 0.490 80 0.20 53
Northern Illinois 4.8% 43 0.03 56 8.8% 19 -0.05 100 3.3% 19 2.7 36 -0.7 69 0.489 81 0.45 44
Western Kentucky -0.5% 78 -0.05 86 -0.4% 80 -0.07 106 1.0% 54 -2.0 93 3.0 30 0.487 82 -0.31 80
Oklahoma State 1.9% 57 0.00 71 -6.1% 104 0.10 17 -1.0% 88 1.2 55 -7.2 123 0.485 83 -0.24 75
Buffalo 6.3% 32 -0.35 125 -0.1% 76 -0.10 114 3.3% 21 3.6 26 -4.6 109 0.480 84 -0.73 96
Bowling Green -3.6% 98 0.00 72 0.2% 71 0.02 59 1.8% 41 2.9 34 -3.9 104 0.480 85 -0.47 86
Oregon State 0.8% 69 -0.04 83 1.9% 58 0.05 41 -0.7% 81 2.3 40 1.7 45 0.477 86 -0.01 63
North Texas -11.8% 126 -0.13 111 -9.9% 119 0.13 4 -2.4% 105 -4.3 112 2.3 37 0.469 87 -1.34 119
Massachusetts -3.8% 100 0.06 49 -4.8% 100 -0.02 83 -1.6% 97 -7.6 124 1.6 48 0.468 88 -0.52 89
Florida International -0.6% 80 -0.26 123 0.3% 70 -0.02 79 -2.3% 103 -0.9 77 8.9 6 0.468 89 -0.59 91
Syracuse -7.6% 115 0.05 51 -8.9% 115 0.02 60 1.7% 43 3.7 24 -0.7 71 0.467 90 -0.67 95
San Jose State -1.1% 84 -0.07 96 -6.9% 110 -0.10 113 0.4% 63 -1.0 79 -7.0 122 0.465 91 -0.95 107
Washington State 0.5% 70 -0.12 105 -2.9% 91 -0.22 126 2.3% 33 -6.1 117 -3.9 105 0.462 92 -0.98 110
North Carolina -3.4% 95 -0.10 100 -5.7% 103 -0.04 93 0.3% 65 -2.9 102 5.9 14 0.458 93 -0.62 93
Purdue -6.3% 110 0.00 73 -3.4% 95 -0.03 87 0.7% 60 -1.3 84 -0.7 70 0.457 94 -0.77 99
Rutgers 1.2% 65 -0.07 98 3.1% 54 -0.03 89 3.3% 22 3.8 22 -6.8 121 0.451 95 -0.36 82
New Mexico -7.7% 116 0.11 35 -9.8% 118 0.06 30 -4.3% 117 3.6 25 0.4 59 0.447 96 -0.59 92
Northwestern -1.9% 86 -0.13 110 1.9% 59 -0.13 121 -1.8% 98 1.3 54 3.7 24 0.445 97 -0.67 94
Colorado 1.3% 62 -0.28 124 0.1% 75 -0.02 81 2.5% 31 0.6 63 -5.6 117 0.444 98 -0.88 103
Indiana 0.8% 68 0.02 68 0.2% 72 0.06 34 -2.9% 111 -1.5 89 -2.5 92 0.443 99 -0.16 70
Hawaii -3.4% 96 -0.16 116 -6.3% 106 0.01 62 -1.5% 95 -6.0 116 1.0 54 0.430 100 -0.96 109
Team Succ. Rt.
Margin
Rk IsoPPP
Margin
Rk RZ SR
Margin
Rk 'New'
STE
Rk Sack
Margin
Rk Exp. TO
Margin
Rk TO
Luck
Rk % of
Pts
Rk 5
Factor
Rk
Wyoming -8.8% 121 0.02 67 -10.0% 120 -0.12 118 -5.8% 123 0.1 68 -0.1 64 0.428 101 -1.30 116
Fresno State -6.5% 112 -0.02 78 -0.5% 81 0.06 31 1.3% 46 0.1 67 -5.1 113 0.425 102 -0.76 98
Illinois -3.9% 101 0.02 61 0.1% 74 -0.04 95 -0.9% 87 -2.4 95 -5.6 118 0.423 103 -0.73 97
Ohio -6.5% 111 -0.06 94 -1.3% 85 -0.05 101 0.2% 67 -6.2 119 2.2 40 0.414 104 -0.89 104
Idaho 0.4% 73 -0.37 126 1.9% 60 0.02 56 -2.8% 110 -9.1 128 1.1 53 0.414 105 -1.13 112
New Mexico State -8.5% 118 -0.05 88 -12.2% 123 -0.21 124 0.3% 64 -7.1 122 -1.9 85 0.409 106 -1.63 124
Army -6.7% 113 0.03 59 -3.1% 92 -0.03 86 -9.6% 127 0.9 57 -0.9 75 0.407 107 -0.96 108
Texas Tech -3.3% 94 0.08 44 -2.2% 88 0.01 61 4.4% 9 -3.0 104 -9.0 128 0.406 108 -0.51 87
UTSA 0.8% 67 -0.18 119 3.4% 51 0.11 13 -0.5% 77 -1.1 81 -0.9 73 0.406 109 -0.37 83
Florida Atlantic -3.5% 97 -0.05 92 -14.9% 125 -0.04 91 3.5% 17 -3.1 106 0.1 62 0.406 110 -0.81 102
Miami (Ohio) -7.0% 114 -0.03 81 -6.4% 107 -0.02 84 -4.8% 119 -4.7 113 4.7 19 0.404 111 -0.93 106
Iowa State -11.5% 125 -0.08 99 -7.5% 111 0.05 38 -0.9% 85 -2.2 94 3.2 26 0.402 112 -1.19 114
UL-Monroe -3.2% 93 -0.13 112 -9.1% 116 -0.05 103 3.0% 25 -1.4 85 2.4 35 0.397 113 -0.80 101
South Florida -11.4% 124 0.09 40 -18.4% 127 -0.01 76 0.7% 60 -1.5 87 1.5 50 0.395 114 -1.08 111
Connecticut -8.7% 119 0.03 58 -6.9% 109 0.05 44 -7.3% 125 -1.2 83 -4.8 112 0.387 115 -1.17 113
Vanderbilt -3.6% 98 -0.13 109 -4.5% 99 -0.11 117 -2.6% 106 -0.6 76 -7.4 124 0.382 116 -1.34 118
Tulane -5.4% 104 -0.01 76 -9.5% 117 -0.21 125 0.9% 57 1.9 44 3.1 28 0.382 117 -0.93 105
Tulsa -6.0% 107 -0.45 128 -6.8% 108 -0.06 105 -1.1% 89 -3.2 107 -1.8 84 0.370 118 -2.04 127
Southern Miss -8.7% 120 -0.16 117 -14.4% 124 -0.10 116 -1.2% 90 -9.0 127 5.0 17 0.367 119 -1.57 123
Georgia State -9.8% 123 -0.05 90 -7.7% 113 -0.01 78 -3.7% 113 -6.1 118 -8.9 127 0.359 120 -1.70 125
UNLV -6.3% 109 -0.15 115 -5.6% 101 -0.04 92 -1.3% 94 -5.4 115 -3.6 101 0.356 121 -1.38 120
Kansas -4.0% 102 -0.12 106 -2.6% 90 -0.38 128 -4.1% 115 -1.9 91 2.9 32 0.337 122 -1.42 121
Old Dominion -5.8% 106 0.11 29 -6.3% 105 -0.18 123 -0.9% 86 2.9 33 -1.9 86 0.333 123 -0.77 100
Wake Forest -13.6% 127 -0.10 101 -0.3% 79 0.17 1 -7.6% 126 0.7 60 -4.7 111 0.333 124 -1.53 122
Kent State -9.4% 122 -0.01 77 -18.1% 126 -0.07 110 -1.9% 100 -3.1 105 3.1 29 0.325 125 -1.30 117
Troy -6.0% 108 -0.18 118 -5.6% 102 -0.12 119 -1.3% 92 -1.9 92 2.9 31 0.311 126 -1.27 115
Eastern Michigan -7.7% 117 -0.25 121 -10.9% 122 -0.04 97 -7.2% 124 -6.2 120 -3.8 103 0.273 127 -2.01 126
SMU -21.4% 128 -0.43 127 -22.3% 128 0.12 8 -5.5% 121 -7.3 123 -5.7 119 0.127 128 -3.46 128

So there's a lot to unpack here; too much, really. (The main takeaway: SMU is more than three standard deviations worse than the average team here.) Like I said, this was just an experiment to see how close I could get, and this gets pretty damn close. So what's next?

1. Keep dialing in. What do you figure are some of the game factors unaccounted for here? This basically explains about 87 percent of a team's percentage of points scored -- what about the other 13 percent? General randomness? Some blind spots with the factors above?

2. Adjust for opponent. Obviously. If this is supposed to lead to some sort of new S&P+ ratings, opponent adjustments are mandatory. I like Marshall more than a lot of folks, but they're probably not the best team in the country, huh?

3. Keep trying to figure out what the hell to do with turnovers. Using sack margin, expected turnover margin, and this turnovers luck figure was actually more effective than I thought, but that's the roughest of rough attempts. There are better ways to figure this out ... I guess. Maybe. Possibly.

4. Isolate success rate in scoring opportunities and non-opportunities. There's some overlap in using success rate and red zone success rate.

5. Play with special teams some more.

This is fun! Right?

This week at SB Nation

Monday
November is here. The college football season has begun.

Tuesday
The Numerical: Don't ever tell Will Muschamp the odds
2014 Advanced box scores
Missouri 20, Kentucky 10: Beyond the box score

Wednesday
Re-projecting the College Football Playoff race using advanced stats
Updated SEC projections through 10 weeks
Updated Big Ten projections through 10 weeks
Updated Big 12 projections through 10 weeks
Updated ACC projections through 10 weeks
Updated Pac-12 projections through 10 weeks

Thursday
Ohio State's chance at decades' worth of Michigan State payback
LSU as Alabama's Playoff spoiler? Now that's terrifying.
College football projections: Week 11 F/+ picks
Bill and Ian overreact over the new playoff rankings

Friday
Kansas State vs. TCU might be the best part of a loaded college football Saturday

Posted by: Bill Connelly on 07 Nov 2014

9 comments, Last at 14 Nov 2014, 6:21pm by Will

Comments

1
by Kal :: Fri, 11/07/2014 - 4:05pm

Might try simply negative yard plays (similar to havoc rate) on the sacks. Sacks don't factor much into, say, GT, but negative plays certainly do. And I bet they cause fumbles and correlate better to interceptions as well. You might even check if penalties matter in this case.

Something I wanted to talk with you at some point is what I've observed as the Baylor effect on S+P. (this is true for teams like Oregon as well, but was really, really true last season when Baylor had something like a 210 passing down rating for S+P). The idea is that there are some teams that are incredibly, absurdly efficient against weak competition but past a certain point that becomes nonpredictive. Blowing out weak opponents is illustrative to a degree, but only a degree. This might be true for blowing out any opponents, period. It goes beyond garbage time removal and simply indicates that when a team is outmatched there is only so much the data can provide that's useful.

8
by Bill Connelly :: Sat, 11/08/2014 - 9:54am

That's certainly something I can look at that, though my immediate response is that, because of the garbage time definitions at hand, if you completely and totally destroy a really bad team, we only end up counting like 25% of the game's plays at most. So Baylor's ratings end up being a mix of 50 plays against Northwestern State and the whole 170 or whatever against WVU. (Plus, dominating those teams to that degree still means something since not every team would dominate to that degree.)

2
by Kal :: Fri, 11/07/2014 - 4:30pm

Also, the big breakthrough that'll probably give about 10% of that predictive value is injuries, but we both know how difficult that is to even guess at.

7
by Bill Connelly :: Sat, 11/08/2014 - 9:51am

Yeah, thanks for the backup on that topic at SBN these last couple of week... :)

3
by Will :: Fri, 11/07/2014 - 6:27pm

Wondering if home field adequately captured, especially since it varies dramatically by team in college football. Playing at UMASS at noon is much easier than playing at Death Valley or at Beaver Stadium at night. I wouldn't be at all surprised if a night game at LSU or Penn State is worth a good 5 points by itself.

Will

5
by Bill Connelly :: Sat, 11/08/2014 - 9:50am

That's kind of zero-sum, though, isn't it? It shows up as a benefit for one team but a detriment for the other?

9
by Will :: Fri, 11/14/2014 - 6:21pm

When I thought about after the fact, I realized this was probably the case. Any effect would also be captured in the per play stats anyway. Would be a cool project to see if S&P+ can actually quantify the various home field advantages, although you would probably have to do five year averages or something, and even longer if you want to see how much more (if any) "night game at Death Valley" is worth over "day game at Death Valley".

Will

4
by ammek :: Sat, 11/08/2014 - 7:13am

Yes, it is fun!

I think some attention has to be paid to the circumstances of each play, in particular the order or sequence in which things happen. A team that converts a lot of third downs, for example, may have a mediocre success rate, but if your model aims to be descriptive rather than predictive, do third-down conversions need to be factored in?

6
by Bill Connelly :: Sat, 11/08/2014 - 9:51am

Certainly possible, though the best offenses are the ones that convert on first and second down. So much of this comes down to leaving yourself margin for error.