Weekly Playoff Probabilites - Final

As always, explanation here, ranking below, thoughts after.

Rank	Team	Agg Rank	POFF Prob	Change
1	Alabama	1	100.00%	0.3%
2	Ohio State	3	96.71%	-0.1%
3	Clemson	4	96.68%	7.7%
4	Washington	5	66.67%	18.6%
5	Michigan	2	15.99%	-6.6%
6	Penn State	14	15.84%	9.4%
7	Oklahoma	8	5.24%	0.4%
8	Western Michigan	22	1.80%	-14.7%
9	West Virginia	19	1.07%	0.3%

1.  After several weeks where it looked like we might have something weird happen, the playoff field will likely be an easy selection of all major conference teams with one loss or fewer.  If that is what happens, then it is unlikely that I change my model at all, beyond some small tweaks.  If Ohio State or Washington get left out, I may have to make some changes, either to SOS sensitivity (if Washington loses out to one of the Big Ten teams with a better SOS) or adding in conference championships and/or head-to-head as a factor (if Ohio State loses out to Penn State). 

2.  The final week of eliminations was a tough one: Oklahoma State, Wisconsin, and Colorado all made it so close to the end.  After finishing with 13 teams in the standings last year, this year sees just 9 teams finish with a good enough record to be considered.  2016 may not have had a ton of chaos at the top of the rankings, but it got pretty crazy after that.

3. If you've been following this closely, you may have noticed that Western Michigan dropped a good deal.  This is because I finally accounted for strength of schedule in the undefeated group-of-five bucket.  Previously, I had been assigning a 2/11 playoff probability for any non-major conference team that finished undefeated (11 teams have done this in the BCS/playoff era, and two of them finished in the top four).  While this didn't normally cause an issue, it was clear that Western Michigan didn't have a better chance of making the playoff than Michigan and/or Penn State.  So, I fit a logit model to the 11 data points I had, much like I had done for the one-loss and two-loss buckets.  Sure, this model isn't as good as the others, as eleven data points can only give you so much info.  That said, it should at least be able to separate the minor conference teams that actually have a chance from those that don't.