So I spent hours and hours this summer putting together a statistical model to attempt to predict individual games and season record for teams. Here is a brief recap of how it did.
The model missed all 8 of the FCS wins over FBS teams. My model isn't really designed for finding which of those FCS "upsets" will happen, since I only include data for FBS schools. In looking back at the actual losses that occurred, my model would have predicted 3 of the 8 games correctly if I had included the FCS data into it.
Excluding FCS games and looking at straight winners, not looking at the point spread, the model went 32-2. The only misses: BYU at Virginia and Colorado State vs. Colorado.
So the model was accurate in predicting the other upsets and close games that occurred over the weekend. It got Clemson over Georgia, LSU over TCU, Ole Miss over Vandy, Washington over Boise State, Fresno State over Rutgers, Cincinnati over Purdue, Northwestern over Cal, Texas Tech over SMU, Western Kentucky over Kentucky, Texas State over Southern Mississippi, UTSA over New Mexico, and Troy over UAB.
If you are in a Pick'Em League, here are my "upset" (or close game) picks for the week:
Boston College over Wake Forest
Miami over Florida
Temple over Houston (Houston better team, but on the road)
Bowling Green over Kent State
Ball State over Army
Utah State over Air Force
Georgia over South Carolina
Duke over Memphis
Indiana over Navy (Navy better team, but on the road: I personally will pick Navy here)
Texas over BYU
Notre Dame over Michigan
UTEP over New Mexico
I haven't actually looked at the FCS matchups yet, I'd put UMass, Akron, Western Michigan, and Georgia State on upset alert. Kansas, Colorado, and Vanderbilt are the 3 BCS teams I think most likely to lose...
Pretty good, Mo. It will be fun to see how your model does over the course of the season. Is there any updating element that takes into account recent performance, or is it only based on past years?
ReplyDeleteI updated it this week with results from week one. There wasn't a ton of movement b/c most teams played nobody in Week One. I'll post the updated rankings today.
ReplyDeleteWithout giving away the farm, how much does recent (read, this year's) performance affect the rankings as opposed to the prior year results.
ReplyDeleteAs a stats guy, I'm interested in your methods :)
E-mail me at the address listed on this site and I'll send you a copy of the rankings. Basically, I assign a pre-season ranking and that holds a lot of weight after week one, the other weights are total wins, total losses, and SOS. Each week as the season goes along, the SOS and record will hold more weight in that week's ranking and the prior week's ranking will hold less weight. Honestly, it's still a work-in-progress in terms of how much the weights will change from week-to-week.
ReplyDelete