≡ Menu

A futile twist on SRS scores for college football teams

Can you imagine Rambo without his knife? Well, I can't imagine not being able to use the SRS.

As a guy who loves the Simple Rating System, the start of the college football season can be frustrating. Until we get a few weeks in, we can’t really use the Simple Rating System to analyze results. A similar problem affects all computer rating systems. Jeff Sagarin notes in his rankings that that for the first few weeks of the season — until the teams are “well connected” — he uses a starting weight to help match the results to our expectations. By “well connected” Sagarin simply means we need the teams to play each other more frequently so we can have more confidence in our results.

So while it’s fun to watch the games, part of me is disappointed that I can’t run all the teams through the simple rating system. If only we could double the sample size of the games played — after six weeks, teams are fairly well connected — we’d be in business.

Well, I thought of a cheat to do just that. Each game has a point spread, a quantitative expectation of how each team will perform according to the best minds in the betting community. So when Florida plays Tennessee, there are two games that are happening. One is the result of the actual game, Florida winning by 17. The other is what we expected to happen, which is Tennessee winning by 3.

It might seem odd to use projected results as inputs into the SRS. And maybe it is. But why not? It’s similar in ways to using preseason projections, and I generally have faith in the betting community. Plus, what else I am supposed to do until a few more weeks.

So here’s what I did. I took every game where the point spread was less than 35 points (under the assumption that games with ridiculous betting lines or that don’t even have a line are just tune ups) and counted it twice. Once for the actual results, and once for the point spread.

Washington, for example, has played 3 games this year. They played San Diego State, were favored by 15, and won by 9 points. They played LSU, were 22.5 point underdogs, and lost by 38 points. And they played Portland State, were 32-point favorites, and won by 39 points. As a result, in six “games” this year, Washington won by an average of 5.8 points. Their strength of schedule was pretty tough, too — iterated, of course — so they will fare pretty well. How does the rest of college football look?

RkTmGMARGSOSSRSTotRecActRec
1LSU230.333.763.92-01-0
2Cincinnati214.347.261.42-01-0
3Florida St2407.747.72-01-0
4Pittsburgh6-2.749.947.22-41-2
5Oregon22521.146.12-01-0
6Youngs St2-2.347.244.91-11-0
7VA Tech40.642.643.23-11-1
8W Virginia230.88.939.62-01-0
9GA Tech48.929.137.92-21-1
10Texas424.312.937.14-02-0
11Georgia21125.136.12-01-0
12Alabama427.87.134.94-02-0
13Washington65.827.933.74-22-1
14Kansas St627.3532.36-03-0
15Oklahoma223.32.725.92-01-0
16Missouri4-3.628.725.12-21-1
17USC49.515.5253-11-1
18Texas Tech631.3-6.724.66-03-0
19Stanford613.6922.65-13-0
20San Diego St44.117.621.72-21-1
21Arizona4714.621.63-12-0
22Arizona St618.42.921.34-22-1
23Fresno St611.39.821.14-22-1
24Oklahoma St413.16.319.43-11-1
25Louisville410.17.617.74-02-0
26Notre Dame613.93.1175-13-0
27Wisconsin69.5716.55-12-1
28UCLA614.32.116.35-13-0
29Florida69.56.616.14-23-0
30S Carolina624.3-8.316.16-03-0
31Ohio State621-5.115.96-03-0
32Texas A&M413.31.915.13-11-1
33Nebraska616.7-1.615.15-12-1
34Oregon St2-1.516.5151-11-0
35Mississippi63.811.2154-22-1
36Virginia62.212.814.94-22-1
37Michigan4-3.11814.92-21-1
38TX Christian216.3-2.713.62-01-0
39Ohio610.2313.15-13-0
40N Carolina615.8-3.612.23-31-2
41North Texas44.47.211.52-21-1
42BYU49.32.211.53-11-1
43Michigan St68.82.210.95-12-1
44Iowa State43.17.810.92-22-0
45Wyoming4-12.823.510.71-30-2
46Northwestrn63.26.9104-23-0
47Utah40.19.89.92-21-1
48Toledo6-0.310.19.83-32-1
49Penn State636.49.43-31-2
50Iowa65.34.19.45-12-1
51Miami (FL)64.54.594-22-1
52Wake Forest6-10.1198.93-32-1
53Marshall6-210.98.92-41-2
54Vanderbilt4-4.413.18.71-30-2
55Baylor221-12.48.62-01-0
56Syracuse6-0.58.98.43-31-2
57Utah State65.82.38.13-32-1
58LA Tech412.5-4.48.14-02-0
59Purdue617.8-10.67.24-22-1
60Tennessee40.85.96.73-11-1
61Tulsa410.9-4.66.23-11-1
62San Jose St69.1-3.164-22-1
63Boise State49-4.44.62-21-1
64California65.4-1.443-31-2
65Boston Col65.6-23.52-41-2
66N Illinois67.8-4.33.44-22-1
67Illinois4-2.45.12.72-21-1
68Clemson415.3-12.62.74-02-0
69TX El Paso6-7.810.52.72-41-2
70Army4-11.112.61.40-40-2
71Air Force48.1-71.12-21-1
72Miss State410.4-9.31.14-02-0
73Central FL610.5-10.10.44-22-1
74Kentucky62.1-1.60.43-31-2
75Nevada4-1.40.8-0.52-21-1
76Arkansas4-1110.4-0.61-30-2
77Houston6-5.75-0.71-50-3
78W Kentucky416-17.1-1.13-12-0
79NW State2-2624.6-1.40-20-1
80Portland St2-35.533.7-1.80-20-1
81Bowling Grn6-5.23.2-1.92-41-2
82N Iowa4-1513-20-40-2
83S Florida68.1-10.4-2.34-22-1
84Kansas62.3-5-2.73-31-2
85Rutgers48.4-11.3-2.93-12-0
86Weber State2-2421.1-2.90-20-1
87Duke4-7.63.7-3.92-21-1
88NC State67.3-11.5-4.34-22-1
89N Dakota St27.5-11.8-4.31.5-0.51-0
90Wash State6-0.8-4.7-5.54-22-1
91Auburn6-2.2-3.4-5.62-41-2
92Missouri St2-3832.3-5.70-20-1
93Central Ark2-21.315-6.20-20-1
94Liberty2-14.37.7-6.60-20-1
95Minnesota611.6-19-7.46-03-0
96Rice6-15.47.2-8.21-51-2
97Stony Brook2-16.88.4-8.30-20-1
98Navy4-2213.2-8.80-40-2
99LA Lafayette4-12.33.2-92-21-1
100Richmond2-2414.9-9.10-20-1
101New Mexico2-33.824.6-9.20-20-1
102N Mex State6-6.2-3.2-9.42-41-2
103Texas State4-21.411.9-9.41-31-1
104Arkansas St4-6-3.5-9.52-21-1
105Connecticut69.5-19.2-9.74-22-1
106W Michigan63.1-14.2-11.12-41-2
107Middle Tenn67.3-18.5-11.15-12-1
108S Mississippi4-13.31.6-11.61-30-2
109McNeese St2-0.5-11.1-11.61-11-0
110E Carolina6-3.3-8.5-11.83-32-1
111Colorado St6-7.8-4-11.81.5-4.51-2
112E Washingtn4-1.1-10.9-121-31-1
113S Methodist6-4.7-7.7-12.42-41-2
114Maryland61.9-14.4-12.43-32-1
115Temple412.1-25.1-12.93-11-1
116Troy6-3.1-9.9-132-41-2
117UNLV6-2.8-10.7-13.51-50-3
118Colorado6-8.3-5.4-13.72-40-3
119LA Monroe4-11-3.1-14.11-31-1
120Central Mich4-3.5-11.6-15.11-31-1
121Florida Intl6-3.8-11.4-15.22-41-2
122Idaho4-9.3-7-16.21-30-2
123Indiana612.4-29.2-16.85-12-1
124Kent State4-3-14.8-17.82-21-1
125Illinois State27.3-25.2-17.91-11-0
126Texas Southern2-29.511.5-180-20-1
127N Arizona4-233.9-19.11-31-1
128Ball State6-6.4-13.1-19.53-32-1
129S Utah4-25.66.1-19.50-40-2
130Miami (OH)6-14.3-5.5-19.72-41-2
131UC Davis2-266-200-20-1
132W Carolina2-298.9-20.10-20-1
133Tulane4-21.91.7-20.20-40-2
134S Dakota St2-17.8-2.7-20.40-20-1
135Wm & Mary2-8-12.4-20.40-20-1
136Fla Atlantic40.1-21.7-21.62-21-1
137UAB4-23.31.6-21.70-40-2
138Memphis6-7.3-14.8-22.11-50-3
139Beth-Cook2-31.39-22.20-20-1
140TN Martin4-14.4-9.3-23.71-31-1
141E Kentucky2-317.2-23.80-20-1
142E Michigan6-15.1-10.1-25.22-40-3
143Maine2-28.83.5-25.20-20-1
144Sacramento State4-14-11.6-25.61-31-1
145Indiana St2-10.5-16.8-27.30-20-1
146Chattanooga2-25.5-2.3-27.80-20-1
147Idaho State2-301.1-28.90-20-1
148App State2-18.3-11.8-300-20-1
149Akron6-4.3-26.6-30.92-41-2
150South Alabama6-5.4-26.1-31.53-31-2
151Wagner2-10.8-21.6-32.40-20-1
152N Hampshire2-25-7.4-32.40-20-1
153Towson2-13.5-19.1-32.60-20-1
154Texas-San Antonio2-1.3-31.5-32.81-11-0
155Elon2-47.313.4-33.80-20-1
156SE Missouri2-19-15.1-34.11-10-1
157S Illinois2-17.3-19.7-370-20-1
158Austin Peay2-34.8-2.3-37.10-20-1
159Villanova2-24.8-12.9-37.70-20-1
160E Illinois2-26.8-11.1-37.80-20-1
161Buffalo227-65.1-38.12-01-0
162U Mass4-27.6-13.3-40.90-40-2
163Nicholls St2-9.8-31.5-41.30-20-1
164Ste F Austin2-34.5-12.4-46.90-20-1
165Morgan St4-30.6-34.5-65.10-40-2

Well, maybe it doesn’t work so well. I thought including the spread data would help smooth things out, but apparently not. Pittsburgh probably stood out to you, as they are after all 1-2. Well they lost to Cincinnati, who is #2 in this system. And they blew out Virginia Tech, who is also pretty good.

But the problem is they lost to Youngstown State. So we know Pittsburgh isn’t that good. But the problem is, the computers don’t know that. Youngstown State played only three games this year — against Pittsburgh, Valparaiso, and Albany. But the games against the latter two schools were omitted, because they are games against Valparaiso and Albany.

But this leaves us with just one game for Youngstown State. So essentially the loss by Pittsburgh to Youngstown State doesn’t hurt the Panthers that much, because all we know about YSU is they beat Pitt. I thought that including point spread data would be the solution: in this case, YSU is 1-1, because they beat Pittsburgh by 14 points but they “lost” to Pittsburgh by 18.5 points. But after giving this some more thought, that is the same thing as saying Pittsburgh beat YSU by 2.25 points, which again means nothing if YSU hasn’t played anyone else.

And since Pittsburgh was rated so highly, Cincinnati looks even better. So the short version is, this thing isn’t really going to work. But we can investigate some things.

Alabama’s score surprised me. Or perhaps better stated, a computer might be surprised to see why we like Alabama. In the preseason polls, Alabama was ranked #2 and Michigan and Arkansas were top-ten teams. What if I told you that Boston College started the season 3-0, with beatdowns over Western Kentucky, Navy and Louisiana Monroe? You probably wouldn’t think that would make BC a national championship contender, but that’s exactly what the SRS sees here. Consider that Alabama has beaten three teams — Michigan, Arkansas and Western Kentucky. The WKU game is thrown out because the point spread was too high, so we only have two games for the Crimson Tide. Well their two opponents have only played two games against FBS schools this year. [1]Technically, Michigan also played Masschussets, an FBS member this year. But that makes my point less exciting! Michigan beat Navy by 6 and Arkansas lost to Louisiana Monroe. According to the SRS, a brand like Alabama beating Michigan and Arkansas is really no different at this point in the year than BC beating Navy and Louisiana Monroe. The size of the blowouts help Alabama, but the computers just see blowouts over the 35th and 79th best teams.

There are other issues. San Diego State blew out Army, lost to Washington, and beat North Dakota (thrown out because there was no point spread). So for SDSU, they look like a really good team for beating Army and then they lost to Washington… so the SRS makes Washington looks really good for beating SDSU…. and then LSU looks really, really good for beating Washington. That’s pretty much how the SRS works.

Again, the key is to get more connectivity between the teams. I tried. Unfortunately, we’re going to have to wait a few more weekss

References

References
1 Technically, Michigan also played Masschussets, an FBS member this year. But that makes my point less exciting!
{ 4 comments }