So, I decided a good test for my yppd (yards per play differential) prediction model would be to use it for confidence picks for the bowl games.
<i>For some background, I’ve been playing with trying to find a way to predict winners to college football games – the idea being that a team’s ability is best measured by how many yards per play they can get relative to their opposition... so it’s not as important if they are a running team or a passing team or a defensive team or an offensive team, but if they can manage to get 1 yard per play more than their opposition, then they are more likely to win. If they get 1 yard per play less than their opposition, they are more likely to lose.
I don’t use straight yppd – I do an adjustment for strength of schedule when using it to predict games – but it’s pretty much just yards per play differential.
Looking at game data for 2009-2010, 81% of winners won the yppd margin – more than any other non-scoring stat (more than passing yards, rushing yards, first downs, etc). In 93% of games, the winner was either even on yppd and turnovers or ‘won’ one or both of yppd /turnovers. Using the yppd model I’ve made, I have been able to ‘predict’ winners to ~75-77% of games.
So, what I’m trying to say – I think this model is a pretty good representation of a team’s ability.</i>
Back to bowling - I entered into yahoo and scored a 509 – or in the 96th percentile. Not bad, right? Makes me thing that this model is actually pretty good. Granted, I only got 68% of the games right, but that’s because there were more close games than in a typical season. The reason I was able to get in the 96% is because the yppd model gives me a view of what games are of higher confidence... which is reflected in how my picks faired.
Splitting my picks into 3 buckets – 1-10, 11-20, 21-35, my record was:
1-10: 3-7
11-20: 7-3
21-35: 14-1
There are some picks where the model predicted a winner that, after watching the game, I was completely lucky (it picked UM over VT... confidence 26). Other games showed how the model works pretty well in seeing things that others don’t (like WV destroying Clemson... confidence 28).
Overall, I’m pretty psyched as the yppd predictor seems to have legs. I’m going to try to fix it up a bit this offseason and create a second scoring metric – we’ll call it the “big play” factor – so things like turnovers, defensive scores, TD:FG ratio.. I’m going to figure out a way to measure them.
So – to bring this back to ND.
ND finished with a 1.3 yppd score, good for #13.
We beat #6-MSU, 59-Pitt, 66-MD, 70-Navy, 71-AF, 85-Wake, 93-PU, 94-BC.
We lost to #8-Stanford, #14-USC, #15-UM, #17-FSU, #47-USF.
I see good news and bad news.
The bad news (which is largely obvious)...
We went 1-4 against relatively even opponents. The yppd scores aren’t perfect, but I think it’s pretty clear that we are in the same ballpark (if you believe the yppd score means anything). An as-expected performance would have us at 2-3 or 3-2. We also should not have lost to USF... we are significantly better and if we played a conservative game, we should have won (I know, not anything new)... but it is very common for a team that is not great to lose a game it shouldn’t. So... an expected outcome to our season would have been 10-3-ish. 8-5 is 2 games off, so I think it is disappointing that we didn’t fair better and I think Kelly clearly needs to do a better job avoiding the collapses (maybe it is a qb thing...)
The good news (which is not as obvious)...
We had a 1.3 yppd score. That is very good. Last year, we were 0.9, the year before 0.3, the year before 0.0. Check out our yppd scores since 2000:
<pre>
2000 -0.2
2001 0
2002 0.7
2003 0.3
2004 0.2
2005 0.8
2006 0.3
2007 -0.9
2008 0
2009 0.3
2010 0.9
2011 1.3
</pre>
Not bad, right? Our 2 year average of 1.1 is better than any other two years by a significant amount (0.55 next best in 05-06). Maybe, just maybe Kelly is putting the whole thing together.
One other interesting thing worth noting – Post Maryland, our YPPD score was much worse this year. We were at 1.7 through Maryland and then 0.4 afterwards (also 1.6 pre-USC, 0.8 USC and later). This is very different than last year, where our final 5 games were at 1.1 vs our first 8 at 0.8.
<pre>
Rank Institution yppd score
1 Alabama 3.0
2 LSU 2.4
3 Oregon 2.2
4 Oklahoma St. 1.8
5 Houston 1.8
6 Michigan St. 1.5
7 Wisconsin 1.5
8 Stanford 1.4
9 South Carolina 1.3
10 Arkansas 1.3
11 Florida 1.3
12 West Virginia 1.3
13 Notre Dame 1.3
14 Southern California 1.3
15 Michigan 1.2
16 Oklahoma 1.2
17 Florida St. 1.2
18 Georgia 1.1
19 Tulsa 1.1
20 Texas A&M 1.1
21 Penn St. 1.0
22 Southern Miss. 1.0
23 Boise St. 1.0
24 SMU 1.0
25 Baylor 0.8
26 Virginia Tech 0.8
27 Missouri 0.8
28 Northern Ill. 0.8
29 North Carolina 0.7
30 Utah St. 0.7
31 Texas 0.7
32 TCU 0.6
33 Nebraska 0.6
34 Tennessee 0.6
35 Toledo 0.6
36 Clemson 0.6
37 Vanderbilt 0.6
38 Auburn 0.5
39 Miami (FL) 0.5
40 California 0.5
41 UCLA 0.5
42 Louisiana Tech 0.5
43 Arkansas St. 0.5
44 Georgia Tech 0.5
45 Mississippi St. 0.4
46 Nevada 0.4
47 South Fla. 0.3
48 Ohio 0.3
49 UCF 0.3
50 Iowa 0.2
51 Washington 0.2
52 BYU 0.2
53 Temple 0.2
54 Illinois 0.2
55 La.-Lafayette 0.1
56 Arizona St. 0.1
57 Cincinnati 0.1
58 Ohio St. 0.1
59 Pittsburgh 0.0
60 Oregon St. 0.0
61 La.-Monroe (0.0)
62 Arizona (0.1)
63 Virginia (0.1)
64 Utah (0.1)
65 FIU (0.1)
66 Maryland (0.1)
67 Marshall (0.2)
68 Minnesota (0.2)
69 Louisville (0.2)
70 Navy (0.2)
71 Air Force (0.3)
72 East Carolina (0.3)
73 Bowling Green (0.3)
74 Kansas St. (0.3)
75 Miami (OH) (0.3)
76 North Carolina St. (0.4)
77 Kent St. (0.4)
78 Rutgers (0.4)
79 UTEP (0.4)
80 North Texas (0.4)
81 San Diego St. (0.4)
82 Syracuse (0.5)
83 Iowa St. (0.5)
84 Western Mich. (0.5)
85 Wake Forest (0.5)
86 Fresno St. (0.6)
87 Central Mich. (0.6)
88 Northwestern (0.6)
89 San Jose St. (0.7)
90 Washington St. (0.7)
91 Ole Miss (0.7)
92 Hawaii (0.8)
93 Purdue (0.8)
94 Boston College (0.8)
95 Eastern Mich. (0.8)
96 New Mexico St. (0.8)
97 Duke (0.9)
98 Texas Tech (0.9)
99 Colorado (0.9)
100 Army (0.9)
101 Troy (1.0)
102 Western Ky. (1.1)
103 Buffalo (1.1)
104 Connecticut (1.1)
105 Rice (1.1)
106 Ball St. (1.2)
107 Kentucky (1.2)
108 Kansas (1.2)
109 Fla. Atlantic (1.2)
110 UAB (1.3)
111 Middle Tenn. (1.3)
112 Wyoming (1.3)
113 Idaho (1.3)
114 Colorado St. (1.5)
115 Indiana (1.5)
116 Tulane (1.8)
117 Akron (1.8)
118 UNLV (1.9)
119 New Mexico (1.9)
120 Memphis (2.3)
121 Texas St. (3.7)
</pre>
