Rank | Team | FPI Rank | CHAMP Prob | POFF Prob | Change |
1 | Clemson | 6 | 20.29% | 88.02% | 13.7% |
2 | Oklahoma | 1 | 23.90% | 51.95% | -0.6% |
3 | Alabama | 2 | 18.96% | 51.95% | 9.7% |
4 | Ohio State | 3 | 17.01% | 51.95% | -0.6% |
5 | Michigan State | 14 | 8.10% | 51.95% | 15.7% |
6 | Iowa | 26 | 3.82% | 51.95% | -14.0% |
7 | Texas Christian | 7 | 1.72% | 7.46% | -0.1% |
8 | Notre Dame | 9 | 1.49% | 7.46% | -0.1% |
9 | Florida State | 10 | 1.49% | 7.46% | -0.1% |
10 | Stanford | 11 | 1.38% | 7.46% | 3.8% |
11 | North Carolina | 15 | 1.15% | 7.46% | -18.6% |
12 | Oklahoma State | 17 | 0.83% | 7.46% | -0.1% |
13 | Northwestern | 53 | 0.14% | 7.46% | -0.1% |
And now let's reflect the successes and failures of this exercise:
THE GOOD
From a high-level perspective, I was quite happy with my model for most of the season. During the time in which anything was still on the table, the output seemed to reflect a reasonable set of expectations as to what would happen. The simplicity of the model may have missed a few subtleties (such as the soul-crushing inevitability of Alabama), but it gave a really good high-level view of the proceedings. Furthermore, it helped me to see the possibilities of upstart teams like Iowa and North Carolina long before the national media picked up on them. Sure, this was more a function of their weak schedules than anything else, but potential wins are potential wins. Overall, I'm pretty happy with how this worked out.
THE BAD
Well this should be obvious. I love the 2015 version of the Irish, but they are NOT making the playoff, and I think we all know which teams are. The simplicity of this model served itself well for most of the season, but broke down a bit at the end, when it was clear what the committee would and wouldn't do. In last week's post, I mentioned adding both something to measure SOS, and something to account for a glut of one-loss teams. After thinking on this for an extra week, my guess is that I will replace the simple bucket percentages (giving odds of making the playoff based on the odds of finishing with 0, 1, or 2 losses) with a regression that looks at expected losses plus SOS. This should make end of season results look better, but we'll see what happens when I play around with this next summer.
Until next season...
Can't argue with this playcall @BarstoolBigCat pic.twitter.com/Q8ay2uCVXQ
— Colin Sexton (@cmsexton3) December 6, 2015
No comments:
Post a Comment