Weekly Playoff Probabilities - Week 3

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

Rank	Team	Agg Rank	POFF Prob	Change
1	Ohio State	2	72.29%	16.4%
2	Alabama	1	64.14%	-20.3%
3	Clemson	5	32.33%	-9.0%
4	Louisville	3	29.35%	13.3%
5	Michigan	4	27.94%	-7.3%
6	Texas A&M	6	20.24%	14.2%
7	Tennessee	12	15.41%	-5.0%
8	Florida State	7	13.66%	-63.0%
9	Washington	8	11.71%	-4.0%
10	Stanford	10	10.40%	7.9%
11	Arkansas	27	7.54%	6.7%
12	Miami (FL)	14	6.84%	5.7%
13	Wisconsin	26	6.38%	4.7%
14	LSU	11	5.95%	-1.0%
15	Baylor	17	5.56%	3.3%
16	Florida	16	5.19%	4.1%
17	Nebraska	22	5.11%	3.5%
18	Georgia	37	4.72%	4.4%
19	Utah	41	4.34%	4.2%
20	West Virginia	33	4.23%	3.4%
21	Michigan State	34	4.22%	4.0%
22	Boise State	25	4.01%	-1.0%
23	Minnesota	54	3.53%	3.5%
24	Western Michigan	44	2.70%	1.5%
25	UCLA	15	2.57%	1.4%
26	San Diego State	52	2.06%	0.7%
27	Penn State	38	1.87%	1.8%
28	Houston	24	1.61%	0.6%
29	Colorado	45	1.52%	1.5%
30	Oklahoma State	18	1.46%	0.9%
31	Maryland	62	1.41%	1.4%
32	Toledo	47	1.36%	0.4%
33	Ole Miss	13	1.32%	-0.9%
34	Indiana	57	1.31%	1.3%
35	South Florida	31	1.19%	1.0%
36	Central Michigan	55	1.15%	0.5%
37	Oregon	32	1.10%	0.8%
38	Pitt	36	1.01%	0.8%
39	North Carolina	20	0.85%	0.6%
40	Arizona State	50	0.79%	0.7%
41	Georgia Tech	46	0.77%	0.7%
42	Army	71	0.74%	0.7%
43	Kansas State	39	0.68%	0.7%
44	Air Force	72	0.66%	0.6%
45	Georgia Southern	82	0.64%	0.6%
46	Navy	65	0.64%	0.6%
47	Memphis	64	0.64%	0.6%
48	USC	23	0.59%	0.0%
49	Texas Tech	51	0.57%	0.6%
50	Virginia Tech	30	0.56%	0.6%
51	TCU	28	0.53%	0.4%
52	Texas	35	0.41%	-1.6%
53	Auburn	19	0.40%	0.0%
54	Wake Forest	67	0.38%	0.4%
55	Iowa	29	0.29%	-1.2%
56	Oklahoma	9	0.29%	-4.0%
57	North Carolina State	43	0.22%	0.2%
58	BYU	42	0.21%	-0.2%
59	Notre Dame	21	0.17%	-3.6%
60	California	49	0.09%	0.1%
61	Mississippi State	40	0.08%	0.1%
62	South Carolina	59	0.04%	0.0%
63	Arizona	60	0.00%	0.0%
64	Purdue	88	0.00%	0.0%
65	Rutgers	89	0.00%	0.0%
66	Oregon State	79	0.00%	0.0%
67	Boston College	74	0.00%	0.0%
68	Missouri	53	0.00%	0.0%
69	Vanderbilt	69	0.00%	0.0%
70	Duke	66	0.00%	0.0%
71	Illinois	80	0.00%	0.0%
72	Syracuse	75	0.00%	0.0%
73	Kansas	106	0.00%	0.0%
74	Northwestern	58	0.00%	0.0%
75	Washington State	48	0.00%	0.0%
76	Kentucky	86	0.00%	0.0%

1.  Alabama and Ohio State are dominating my model.  In fact, when I ran the numbers they were dominating too much.  I absolutely think the Buckeyes and Tide are great teams with very good chances to do well against strong schedules, but virtually no one should ever have playoff odds greater than 80% three weeks into the season.  So I made some small modifications to how the final number is calculated.  I have more details in point #4, but to summarize: Instead of awarding the majority of "bonus points" to the very best teams, I came up with a way to add bonus points primarily to the power conference teams in the middle of the pack.  Thus, the teens are now filled with teams that got a bit of a bump from this adjustment.  I will likely make a more permanent change to how I calculate the rankings in the offseason (again, see #4), but I feel like this is a good compromise for the time being.

2.  This was a surprisingly quiet week for eliminations (in part because a lot of major conference teams played other major conference teams) as we only lost seven teams.  Goodbye Southern Mississippi, Cincinnati, Marshall, East Carolina, Texas State, Virginia, and Iowa State.  Those last two in particular are the reason the term "Year Zero" was invented.  Bronco Mendenhall and Matt Campbell should be able to do good things, but not this year.

3.  The big move of the week is the fall of the ACC, which is natural as one of their top teams had to lose.  The SEC and Big Ten are more than happy to take over as the conferences in the driver's seat.  The adjustment I mentioned in point #1 does a couple of things here:  One, it boosts a bunch of non-Alabama SEC teams, so the SEC stays ahead of the Big Ten, even though it doesn't boast a top two that is quite as strong.  Two, it rescues the Big 12 from falling to seventh in these rankings, which I think would have been a bit of an exaggeration of their plight.  Sure, it's unlikely that any one team runs the table in the Big 12 and has a strong enough case to make the playoff.  But, it's certainly more likely than the MAC or Mountain West getting a bid. 

Conference	Exp Playoff Teams
SEC	1.250
B10	1.243
ACC	0.860
P12	0.331
B12	0.137
MWC	0.067
MAC	0.052
AMER	0.041
IND	0.011
SB	0.006

4.  Details of the small change:  As I have mentioned before, my method is to simulate a bunch of seasons (10000 of them to be precise), see how often teams end up in one of four buckets (0, 1, or 2-loss teams from major conferences, and undefeated Group-of-5ers) and what their SOS is likely to be, and then calculate their playoff odds based on that.  This is fine and dandy in general, but the problem is that I am using current ratings to predict future game odds, and ratings early in the season are necessarily more clumped together than final ratings.  This means that my system generally predicts something like 0.5 undefeated major conference teams, when in reality there are almost always 1 or 2 at season's end.  Thus, the sum of playoff probabilities often tends toward the 2.5-3 range, as we're necessarily underestimating the number of truly dominant teams.  My solution to this in the past has been to simply multiply everyone's odds by 4/that number.  This usually works, except when there are a couple dominant teams.  Unfortunately, we do have dominant teams this year, and Alabama and Ohio State were both pushing 100% with this method.  While it's pretty likely than one or both eventually make it, we shouldn't be that sure of it (see 2015 Ohio State).

The real solution to this is to calculate playoff odds for each of the 10000 seasons individually, and then add those up for the final probabilities.  However, this would take some coding and even more QAing, which isn't easy to jam in during the season.  So, what I ended up doing was using my good friend Solver to find an even amount of 0, 1, and 2-loss seasons to add to all of the applicable teams, such that the playoff odds all sum to 4.  This will lead to a little volatility the next couple of weeks, but once we get rid of the superfluous major conference unbeaten teams (cough cough Georgia), this should make for a decent approximation of reality.


Week 4 Preview

Home	Away	Home Win Prob	Playoff Teams Lost
Arkansas	Texas A&M	29.4%	0.038
Tennessee	Florida	62.9%	0.030
South Florida	Florida State	34.9%	0.026
Georgia Tech	Clemson	22.5%	0.026
UCLA	Stanford	49.2%	0.024
Michigan	Penn State	87.7%	0.020
Michigan State	Wisconsin	53.1%	0.018
Auburn	LSU	45.1%	0.016
Ole Miss	Georgia	77.5%	0.015
Baylor	Oklahoma State	60.1%	0.012

Week 4 lacks the sizzle of 1 and 3, but makes up for it with fantastic depth.  None of the very top teams will be challenged much more than Clemson's mildly tricky road game against GT, but we'll get a fantastic narrowing down of the mid-level contenders.  Arkansas-A&M will probably be the most fun of all these games, but when half of them are virtual coin tosses, you're in store for a full day of insanity.