Weekly Playoff Probabilities - Final

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

Rank	Team	Agg Rank	POFF Prob
1	Georgia	5	93.58%
2	Clemson	4	91.36%
3	Alabama	1	75.24%
4	Wisconsin	8	54.38%
5	Oklahoma	7	37.85%
6	Ohio State	2	19.17%
7	Miami (FL)	16	9.07%
8	Penn State	3	8.49%
9	USC	17	5.11%
10	Central Florida	14	4.54%
11	Washington	6	1.21%

For the first three years of the Playoff, the top four teams in my model lined up with the top four teams in the committee's rankings.  Thanks to the oddly non-chaotic chaos of 2017, that streak will not extend to four.  Wisconsin isn't going to make the Playoff with zero "quality" wins on their resume, and Oklahoma is a virtual lock.  Still, this means the likely playoff field is #1, #2, #3, and #5 in my model, so I don't consider that a total loss.  Yes, it's possible that Ohio State beats out Alabama for the final spot, but I tend to think that the argument for Bama>OSU is simply the high-falutin' version of last year's argument for Washington>PSU.  We shall see.

Let's take a minute to break down why Wisconsin is ahead of Oklahoma in my model.  Since there are no games remaining, it all comes down to SOS.  While, it is generally accepted that the Sooners' schedule has been superior to the Badgers', the baseline SOS that I use* thinks the opposite is true, with Wisconsin pulling ahead .537 to .521.  There are two reasons for this discrepancy.  First, Oklahoma has three really bad teams on it's schedule.  UTEP, Kansas, and Baylor went a combined 1-34 against FBS competition, which necessarily drags down Oklahoma's SOS.  Second, as I mentioned before, the Big 12 was rather unimpressive in the non-conference slate, with almost everyone outside of the top three teams performing poorly in their biggest matchups.  This has the effect of compounding those losses in conference interplay, which drags down Oklahoma's SOS a bit.

*(2 x Opponent's record + 1* Opponents' Opponent's record)/3, which is the general definition the NCAA uses for the basketball selection process

What does this mean for the future?  Well, it doesn't mean that I shut everything down just yet.  After all, one slightly odd result does not invalidate the entirety of the model.  But it does mean I may need to investigate using slightly different measures of schedule strength.  My original motivation behind SOS estimation was simplicity, and while that may still be the best guiding principal, it might not best reflect the committee's undying focus on big games.  My model currently thinks that it matters that Wisconsin's worst opponents (Illinois, Maryland, BYU) were better that Oklahoma's, but in reality it almost certainly doesn't.   How do I account for this?  Should I even worry about it?  I'm not sure of the answer to these questions.  But I'll reflect on it in the offseason and go from there, just like a certain team that got blown out twice in their last three games.