/cdn.vox-cdn.com/uploads/chorus_image/image/41260378/456677954.0.jpg)
Biggest M&B yet? Sure, why not.
Models and Bottles Refreshers!
Models and Bottles Refreshers!
Models
I'm always making tiny little adjustments to the models during the season, largely because I want to experiment and see if tinkering will improve performance, or have the output data make a little more sense.
My timing is consistently awful, considering we find ourselves in the throws of an Evanston-Dinkytown statistical call-to-arms. By virtue of personal character flaw (or benefit, depending on your purview), I was prompted by suspicious results to go back and review the M&B methodology.
Non-techinical synopsis: the calculations I used were appropriately constrictive given the early season results (don't want to get too high or too low on a squad), though as more data became available, my model required a little flexibility to account for weekly fluctuations. I switched to a Bayesian approach this year because I was hoping to see a point of convergence between preseason, early season and mature season results, and I believe I have accomplished that by stumbling into a set of conditionals that makes sense.
This addendum to M&B has a significant impact to the Top 25, which will no doubt anger #FSUTwitter and require some additional explanation. New versus old ranks are presented for your consideration, which is at the very least interesting given the results of #CHAOSWEEK.
Rank | Team | BayesianProb | Delta from Last Week | Strict Model Rank | Model Diff. |
1 | Baylor | .956 | 1 | 1 | 0 |
2 | Mississippi | .952 | -1 | 9 | 7 |
3 | Auburn | .933 | 3 | 2 | -1 |
4 | TCU | .922 | 3 | 17 | 13 |
5 | Mississippi State | .920 | 3 | 8 | 3 |
6 | Oklahoma | .918 | -2 | 7 | 1 |
7 | Texas A&M | .892 | -4 | 6 | -1 |
8 | Alabama | .888 | 1 | 24 | 16 |
9 | Notre Dame | .880 | 1 | 11 | 2 |
10 | Ohio State | .867 | 11 | 18 | 8 |
11 | Michigan State | .867 | 0 | 15 | 4 |
12 | Memphis | .860 | 10 | 26 | 14 |
13 | Georgia | .857 | 5 | 13 | 0 |
14 | Stanford | .856 | -1 | 10 | -4 |
15 | Louisville | .850 | 11 | 21 | 6 |
16 | Arkansas | .848 | 0 | 36 | 20 |
17 | Arizona | .844 | 6 | 23 | 6 |
18 | Kansas State | .842 | 7 | 43 | 25 |
19 | West Virginia | .830 | 0 | 28 | 9 |
20 | Florida State | .830 | 12 | 4 | -16 |
21 | Oregon | .822 | -7 | 3 | -18 |
22 | Nebraska | .817 | -2 | 32 | 10 |
23 | LSU | .816 | -18 | 5 | -18 |
24 | Utah | .812 | 6 | 29 | 5 |
25 | Oklahoma State | .802 | 11 | 22 | -3 |
46 | Minnesota | .709 | -1 | 40 | -6 |
53 | Northwestern | .675 | 7 | 71 | 18 |
The strict model had the benefit of effectively discriminating between teams that were good and bad through the first two games. This serves well through the non-conference schedule but there is a point at which the between-game variation stabilizes and that becomes a better method of generalizing future results. Additionally, the strict model favors consistency among teams that were really good through the early season (especially the first two games) and equally punishes crappy teams that were consistently awful. What the strict model doesn't do all that well is account for variation after the first few games that would indicate an upward or downward trend. It is slow to account for changes in relative team strength in-between the elite and the dregs, which is a notable flaw.
The flexible model becomes a factor as the number of games played by a team increases. With almost half the season in the books for most teams, the stabilization of performance becomes a statistical property and we can "trust" the in-season, week-to-week results with more confidence. In the mean time, there are noticeable differences in estimation of team strength between the models that adds a bit of intrigue to the weekly prognostication:
Flex Model | Strict Model | |
Gopher Win Probability | 54.0% | 66.4% |
Predicted Margin of Victory | 1.59 | 6.79 |
Current Vegas Spread | -4 | |
Current Vegas Over/Under | 42 |
We're still favored, though more in line with bookie expectations. The flex model accounts for more of Northwestern's improvement over the last two weeks than the strict model, for the reasons I described above.
Note the over/under. Looks like we're in store for another uncomfortably close game against the Cats. Yay.
Flex M&B | F/+ | Strict M&B | |
Mean B1G Wins | 4.59 | 4.95 | 4.90 |
Frequency of 3+ B1G Wins | 96.0% | 98.2% | 97.0% |
Freq. of 5+ B1G Wins | 53.0% | 64.3% | 63.0% |
Freq. of 7+ B1G Wins | 6.0% | 9.5% | 9.0% |
The biggest difference between flex and F/+ projections? How each views Purdue (yes, Purdue), Northwestern and Iowa.
Team | M&B Prob | F/+ Prob |
Ohio State | 27.1% | 32.5% |
Nebraska | 35.2% | 30.2% |
Wisconsin | 48.1% | 37.6% |
Iowa | 52.6% | 70.1% |
Northwestern | 54.0% | 65.0% |
Purdue | 63.1% | 84.8% |
Illinois | 79.5% | 74.4% |
Purdue is also somehow better, offset by LOL Tim Beckman.
To prove that I can do more than just predict Gopher games, here's the Week 7 slate of M&B predictions. Enjoy in moderation.
Visit Team | Home Team | Home Winning % | Expected Margin |
Miami (OH) | Akron | 74.9% | 10.91 |
Southern California | Arizona | 64.2% | 5.85 |
Alabama | Arkansas | 41.3% | -3.53 |
Rice | Army | 27.5% | -9.68 |
Western Mich. | Ball St. | 27.4% | -9.73 |
TCU | Baylor | 64.7% | 6.05 |
Washington | California | 55.5% | 2.19 |
Louisville | Clemson | 39.9% | -4.11 |
Buffalo | Eastern Mich. | 14.6% | -17.68 |
LSU | Florida | 44% | -2.42 |
Duke | Georgia Tech | 57.6% | 3.07 |
Wyoming | Hawaii | 66% | 6.61 |
Indiana | Iowa | 51% | 0.42 |
Toledo | Iowa St. | 48.2% | -0.73 |
Oklahoma St. | Kansas | 14.6% | -17.68 |
Massachusetts | Kent St. | 62% | 4.89 |
La.-Monroe | Kentucky | 89.3% | 21.18 |
Middle Tenn. | Marshall | 75.3% | 11.17 |
Houston | Memphis | 80.8% | 14.35 |
Cincinnati | Miami (FL) | 63.1% | 5.38 |
Penn St. | Michigan | 45.2% | -1.93 |
Northwestern | Minnesota | 54% | 1.59 |
Auburn | Mississippi St. | 45.3% | -1.88 |
Georgia | Missouri | 31.9% | -7.6 |
Colorado St. | Nevada | 30.4% | -8.28 |
San Diego St. | New Mexico | 54.6% | 1.84 |
Boston College | North Carolina St. | 25.5% | -10.71 |
Central Mich. | Northern Ill. | 43.5% | -2.6 |
North Carolina | Notre Dame | 91% | 23.16 |
Bowling Green | Ohio | 39.7% | -4.17 |
Michigan St. | Purdue | 17.9% | -15.21 |
East Carolina | South Fla. | 19.9% | -13.94 |
Washington St. | Stanford | 73.3% | 10.09 |
Florida St. | Syracuse | 19.2% | -14.36 |
Tulsa | Temple | 91.3% | 23.54 |
Ole Miss | Texas A&M | 29.2% | -8.84 |
La.-Lafayette | Texas St. | 71.2% | 9.07 |
West Virginia | Texas Tech | 9.6% | -22.43 |
New Mexico St. | Troy | 33.2% | -6.97 |
UConn | Tulane | 52.5% | 0.99 |
North Texas | UAB | 54.7% | 1.87 |
Oregon | UCLA | 44.1% | -2.39 |
Fresno St. | UNLV | 14.3% | -17.89 |
Air Force | Utah St. | 48% | -0.81 |
Old Dominion | UTEP | 61.7% | 4.75 |
Illinois | Wisconsin | 80.7% | 14.33 |
Bottles
Minnesota and Northwestern are about to play a critically important game that will put the winner in the driver's seat to win the Big Ten West.
/faints
//reads again
///dies after fainting again
The Gophers are continually and more consistently playing in games with something real on the line, which is all you really need to know about the job Jerry Kill and his staff have done. In four short years, he has taken a program left for dead and turned them into serious contenders in the division.
Only in our most lucid of dreams could this fanbase have envisioned such a scenario back in 2010.
Every season since his first, Kill has treated us to a new "biggest game in years." Syracuse 2012, alien as it sounds and equally odd to type, was the first sellout since 2010. The stakes were high for Iowa and Wisconsin last season. This Saturday though, is on a whole other level. A victory will put them in a space not occupied since 2003: in control of their own destiny with a realistic shot at some hardware.
Those words have only been written a few times about the Gophers in the modern era, which makes this week's featured bottle all the more appropriate: Few Spirits Rye. Whiskey Advocate's 2013 Craft Whiskey of the Year, Few is located - ironically enough - in Evanston, Illinois.