Finding the upset: week 11

2-3 last week, sorry for no post. Whatever, no one reads these. 19-14 on the year. Let’s get rolling.

We only have to deal with favorites this week. Top 25 teams included. Win expectancy levels:

• 50-59% (1-1 last week, 5-5 overall): #16 West Virginia (at Texas) (52.3%), #15 Utah (at Arizona State) (59%)
• 60-69% (2-0 last week, 12-3 overall): #8 Texas A&M (vs. Ole Miss) (65.4%), #19 Nebraska (vs. Minnesota) (66.6%)
• 70-79% (5-2 last week, 18-5 overall): #24 LSU (at #25 Arkansas) (70.3%), #13 Oklahoma State (vs. Texas Tech) (74.3%), #10 Penn State (at Indiana) (74.9%), #17 UNC (at Duke) (75.3%), #4 Washington (vs. #20 USC) (75.4%), #14 Virginia Tech (vs. Georgia Tech) (79.2%), #11 Oklahoma (vs. Baylor) (79.5%)
• 80-89% (5-0 last week, 17-6 overall): #9 Auburn (at Georgia) (81.4%), #23 Washington State (vs. California) (83.7%), #12 Colorado (at Arizona) (84.4%), #22 Boise State (at Hawaii) (87.9%)
• 90-99% (6-0 last week, 50-1 overall): #2 Clemson (vs. Pitt) (90.8%), #3 Michigan (at Iowa) (91.3%), #21 Western Michigan (at Kent State) (92.1%), #18 Florida State (vs. Boston College) (92.8%), #7 Wisconsin (vs. Illinois) (94.1%), #6 Louisville (vs. Wake Forest) (96.8%), #1 Alabama (vs. Mississippi State) (97%)

You can find the usual expected losses/undefeated likelihood numbers in the image above.

There’s about 4.4 anticipated losses from Top 25 favorites this week, including two at the 70-79% range. We already have one with Duke over #17 UNC, and for the second one, I’m going against my anti-Washington/LSU instincts. The most powerful indicator of 70-79% upsets the last few weeks has been Atomic Football, which was the lowest model on both Texas A&M and Baylor last week. (Perhaps they hate Texas schools.) They suggest that both Oklahoma State losing to Texas Tech and Penn State to Indiana are worth a look. On the other hand, my spreadsheet suggested a -13.5 point spread for Penn State and Indiana. It’s at -7, which feels both bizarre and an indicator of something to come. (Oklahoma State’s point spread is about 2 points higher than I would’ve anticipated. The other notable spread: LSU -7, around 3 points lower than anticipated.)

I feel pretty comfortable in stating that one of these two will be your other likely 70-79% upset this weekend. I’m personally rolling with Indiana over #10 Penn State because I believe it’s the safest bet.

Time to find two more. The easy one now is Texas over #16 West Virginia in yet another Charlie Strong late-season special to top off our 50-59% range. Now, a disclaimer: from the 50-79% range, we have 3.278 expected losses. This means we need to find the 1.127 in the 80-99% range. It feels weird to go out on a limb and do this, but let’s find one. This feels gross, too, but why not: there’s a 9% greater chance the 80-89% group doesn’t go undefeated compared to 90-99%, so that’s the direction I’m going in. We’re missing one expected loss from the Top 6-15, so that narrows our decision down to #9 Auburn losing to Georgia or #12 Colorado losing to Arizona.

Why not go with a rivalry game with weird energy on both ends? Auburn needed a second-half appearance by their supposedly injured quarterback to defeat Vanderbilt, and Georgia, odd as it may seem, has heavily improved on defense over the last month, posting the 7th-best yards per play rate in October. This is the best defense Auburn will have played since an 18-13 win over LSU where they needed LSU’s final play to be about 0.3 seconds late to win. I’m liking this more and more, or maybe I’m just stupid: Georgia over #9 Auburn.

This week’s picks: Duke over #17 UNC, Texas over #16 West Virginia, Indiana over #10 Penn State, and Georgia over #9 Auburn.

 

Finding the upset: week 9

I took last week off and don’t have much time to write this week’s post in great detail. We’ve reorganized the metrics, dropping S&P+ and adding Atomic Football and the Vegas lines, with win percentages brought to you by S&P+. Here’s the spreadsheet:

The non-Top 25 v. Top 25 games come first. There’s 14 of these this week and around 3.33 expected upsets, so we’ll try to find those. Win probabilities for each category:

• 30-39% (0-1 overall): #22 Navy (at USF) (30.8%)
• 50-59% (2-3 overall): #15 Auburn (at Ole Miss) (54.7%), #25 Virginia Tech (at Pittsburgh) (56.2%), #10 West Virginia (at Oklahoma State) (56.4%)
• 60-69% (9-2 overall): #8 Baylor (at Texas) (61.8%)
• 70-79% (10-3 overall): #14 Florida (vs. Georgia) (70.2%)
• 80-89% (11-4 overall): #13 Boise State (at Wyoming) (82.7%), #24 Penn State (at Purdue) (83.2%), #18 Tennessee (at South Carolina) (84.6%)
• 90-99% (39-1 overall): #6 Ohio State (vs. Northwestern) (93.3%), #2 Michigan (at Michigan State) (95.3%), #5 Louisville (at Virginia) (95.7%), #16 Oklahoma (vs. Kansas) (98.7%), #9 Texas A&M (vs. New Mexico State) (99.7%)

And the likelihoods for each category:

• 30-39%: 0.308 of 1 (30.8% chance of 1-0)
• 50-59%: 1.673 of 3 (17.4% chance of 3-0)
• 60-69%: 0.618 of 1 (61.8% chance of 1-0)
• 70-79%: 0.702 of 1 (70.2% chance of 1-0)
• 80-89%: 2.505 of 3 (58.2% chance of 3-0)
• 90-99%: 4.826 of 5 (83.7% chance of 5-0)

First off, USF over #22 Navy seems pretty obvious. It looks like we’ll get one from the 50-59% range (which would be appropriate to get them to 4-4 on my metrics), so let’s take Ole Miss over #15 Auburn – they’re the best underdog, they aren’t 3-10 as a home dog of 7 points or less like Mike Gundy and Oklahoma State are, and…well, do you trust Pittsburgh? Ever?

Anyway, the third upset is a struggle to find. I could see any of the picks from 60-89% as realistic, and four of the five games are on the road. Using the metrics with Top 25 included (of which the screenshot is missing the Florida game in its calculations), let’s rule out 80-89%, though as a whole per Bill Connelly’s metrics, they’re running slightly ahead of expectation as well. (If you’re looking for a real shocker, don’t rule out Purdue over #24 Penn State.)

With that in mind, there’s less of a chance that the 70-79% brigade goes undefeated than the 60-69% group does. Let’s go with that and get to the 60-69% group at an appropriate time…say, in the Top 25 v. Top 25 section. Anyway, I have some questions about Florida, such as: what’s their best win? Are we sure they’re good? Does either team have an offense? In a game that’s likely to be low-scoring, it benefits the underdog because the game will likely never be out of hand. They’ll always have a chance to strike. Georgia may have an idiot at offensive coordinator, but they also have Isaiah McKenzie, Nick Chubb, and Jacob Eason. Plus, the favorite (UF -7.5) has lost three of the last four in this series, straight up. Plus plus, the favorite hasn’t covered the point spread since 2011. Let’s try it out: Georgia over #14 Florida.

Now, to the Top 25 games.

• 20-29% (1-3 this season): #17 Utah (vs. #4 Washington) (27.3%), #7 Nebraska (at #11 Wisconsin) (29.5%)
• 30-39%: #13 Florida State (vs. #3 Clemson) (38.2%)
• 60-69% (1-1 overall): #3 Clemson (at #13 Florida State) (61.8%)
• 70-79% (5-0 overall): #11 Wisconsin (vs. #7 Nebraska) (70.5%), #4 Washington (at #17 Utah) (72.7%)

I see two obvious picks, the first of which is #11 Wisconsin over #7 Nebraska. This isn’t even an upset, sadly.

Now, our real semi-shocker: #13 Florida State over #3 Clemson. Why? 60-69% favorites in our system this week have just a 39.3% chance of going undefeated. Clemson is a slightly more flimsy favorite than Baylor. While I’d love to take Texas in an absolute must-win spot for Charlie Strong, there is zero reason at this point in time to do that. God, this is a stressful week.

Our record on the year: 15-8 (68.2%), which is 47% ahead of the normal Las Vegas underdog win rate. This week’s picks: USF over #22 Navy, Ole Miss over #15 Auburn, Georgia over #14 Florida, and #13 Florida State over #3 Clemson.

NBA League Pass TV Guide (Week One)

As you’ll note simply by the tagline on this blog – “170 days of NBALP” – there are few things in life I love more than NBA League Pass. There are so many games and so many wonderful things you can do with it, along with so many reasons to own it: #LeaguePassAlert, any time Russell Westbrook has 35/9/9 in the third quarter, any time LeBron James has scored Cleveland’s last 13 points, any time Golden State runs out the greatest lineup in NBA history, and any time the 76ers try to figure out the league’s first five-center lineup.

Anyway, this is a nice way of saying that I and most other NBA fans live for Zach Lowe’s yearly League Pass Rankings. He misses on some and hits on others, but he’s largely right because he watches more of this sport than anyone else, and that includes scouts paid by teams to do so.

I toyed with the idea of doing this last year, but I figure this is the most quantifiable way to give recommended viewing games for all basketball junkies across the universe. Below is a spreadsheet with composite ratings based on both teams playing for each game in the first week of the NBA season. I’ve highlighted the highest-rated game or games for each day, along with a table on the front end showing Lowe’s ratings from best to worst. This should give you an idea of what games are the most watchable past opening night, and I’ll be doing this weekly throughout the season with tweaks to the rankings as need be. Those tweaks will be based on injuries, surprises, disappointments, and much more.

Here’s your first Weekly TV Guide for NBA League Pass of the 2016-17 season:

Finding the upset: week 7

Another great week. Last week’s picks, both Top 25 v. non-Top 25 and Top 25 v. Top 25, went 4-0. That’s the second time this season we’ve put up an undefeated week, and my system is now 14-5 on the year. I have no idea if anyone reads these and they probably don’t, but my hope at the beginning of the season working on this was to simply get to a range of 55-60% by the end of the season, which would signal success to me. I’m currently at 73.7% with seven weeks of football left and feel good about my odds.

Last week’s picks were USC over #21 Colorado (W 21-17), LSU over #18 Florida (not played), and Washington State over #15 Stanford (W 42-16). The Top 25 vs. Top 25 upset picks were #23 Florida State over #10 Miami (FL) (W 20-19) and #25 Virginia Tech over #17 North Carolina (W 34-3). Pretty good!

Now, on to this week’s numbers, Top 25 v. non-Top 25 only to start…

As usual, the top number is the number of total wins, the bottom the likelihood that everyone in this group goes undefeated with an absolute of 1. The number of games in each section this week:

Top 1-5: 1
Top 6-15: 6
Top 16-25: 7

There’s 14 games this week featuring a Top 25 team versus an unranked opponent. If you remove the three Top 25 v. Top 25 games from the above ranking, that gives us about 2.657 expected upsets, which we’ll round up to three. (Also one Top 25 v. Top 25 upset, which we will read into later.)

Win probabilities for each category:

• 40-49% (0-0 last week, 2-3 this season): #20 West Virginia (at Texas Tech) (46.6%)
• 60-69% (1-0 last week, 8-2 this season): #10 Nebraska (at Indiana) (67.5%)
• 70-79% (2-1 last week, 8-2 this season): #16 Miami (FL) (vs. North Carolina) (70.4%), #21 Utah (at Oregon State) (73.4%), #24 Western Michigan (at Akron) (75.8%)
• 80-89% (2-1 last week, 9-3 this season): #19 Oklahoma (vs. Kansas State) (80.5%), #18 Florida (vs. Missouri) (80.6%), #17 Virginia Tech (at Syracuse) (80.7%)
• 90-99% (2-0 last week, 33-1 this season): #3 Clemson (vs. NC State) (90.3%), #14 Houston (vs. Tulsa) (91%), #15 Florida State (vs. Wake Forest) (91.7%), #16 Boise State (vs. Colorado State) (93.5%), #7 Louisville (vs. Duke) (95.9%), #11 Baylor (vs. Kansas) (96.3%)

Now, the expected wins for each category:

• 40-49%: 0.466 of 1 (46.6% chance of 1-0)
• 60-69%: 0.675 of 1 (67.5% chance of 1-0)
• 70-79%: 2.196 of 3 (39.2% chance of 3-0)
• 80-89%: 2.418 of 3 (52.4% chance of 3-0)
• 90-99%: 5.586 of 6 (65% chance of 6-0)

There’s two obvious places of attack here: the 40-49% and 70-79% ranges. Because there’s just one game in the former, we’ll go ahead and lock in Texas Tech over #20 West Virginia. Finding one in the 70-79% range is a bit more difficult because I don’t like any of the underdogs or their chances, but in instances where it’s relatively close, always take the best underdog by the computer rankings. That would be North Carolina over #16 Miami (FL).

As usual, the third will be a slight stretch. Let’s go to the group with the third-lowest chance of an undefeated record – 80-89%. All of these teams have almost exactly the same odds of winning, which is very rare, but we had the exact same thing happen last week. However, this group is performing well below their expected capabilities. Let’s fade this group and go with Indiana over #10 Nebraska.

A side note: this may be the week we get an earth-shatterer of an upset. It’s been five weeks since the last 90%+ team lost, and they’re about one game ahead of expectation. A 65% chance of an undefeated record is also the lowest this group has been since that week. I won’t be picking it as it doesn’t fit with the metrics, but there’s a real chance someone leaves this weekend totally stunned. There are two underdogs that stand out from the pack ratings-wise: NC State and Wake Forest. Perhaps the ACC race takes a shocking turn this weekend. We’ll see.

Now, on to the Top 25 v. Top 25 games. Same thing with probabilities:

• 60-69% (1-0 last week, 1-0 overall): #12 Ole Miss (at #22 Arkansas) (68.9%)
• 70-79% (3-0 overall): #1 Alabama (at #9 Tennessee) (74.9%), #2 Ohio State (at #8 Wisconsin) (76.2%)

And expected wins:

• 60-69%: 0.689 of 1 (68.9% chance of 1-0)
• 70-79%: 1.511 of 2 (57.1% chance of 2-0)

In total, there’s about 0.8 expected losses from Top 25 favorites on Saturday, which can be rounded to a whole one loss. As with above, we’re going to take the best underdog, considering that the three of these teams are relatively close in odds to win. Here’s how the Massey Composite, which takes in 102 different computer rating systems, says these underdogs rank:

• Tennessee (#8)
• Wisconsin (#9)
• Arkansas (#31)

Well…that’s certainly something. It looks to me like we can rule Arkansas out, which is fine, because it still befuddles me that Arkansas is ranked to begin with. It’s now down to Tennessee or Wisconsin, who at #8 and #9 are more or less interchangeable. My own bias tells me not to take Tennessee, because I know there’s a small chance it actually happens. Eff it, though: I’m standing by my best underdog rule. #9 Tennessee over #1 Alabama.

The Ringer is bad. Here’s why

I considered doing this as a long Twitter thread, but that would be just as bad and disappointing as The Ringer has been thus far. Also, this is only partially inspired by this:

Moving on.

1. The site’s formula for success is inherently wrong.

Bill Simmons doesn’t seem to understand very well that the following site formula:

=SUM(ESPN+Hollywood Reporter+The Verge)/3

Simply isn’t working. Indeed, all three areas are very average, with the best part of the site typically coming in varying sports sections – what Simmons is most well-known for. This sort of weird stew that attempts to make readers think there’s a natural crossover between New York/Hollywood/Silicon Valley is a bit ridiculous. The Ringer is a great example of a restaurant theory I hold: focus on what your niche is and what you do well. By aiming to be the Applebee’s of websites, you’re a jack of several trades and a master of none.

At no point during Grantland‘s tenure was the front page as watered down with entirely useless stories as this sample from today’s front page is.

Thanks for letting me know that some of the clown photos aren’t real, Molly. How would I ever have known?

2. Someone’s convinced Simmons that tech coverage is what was missing from Grantland.

Listening to Simmons attempt half-hearted interest in technology whenever it’s mentioned on one of his podcasts is more sad than frustrating. I’ve yet to read an article on The Ringer centered around tech that didn’t sound like one of the following:

• An edited press release
• 300-400 words that might as well have been five Tweets
• A Verge or CNet castoff

3. The Ringer is riding Shea Serrano until he dies out.

I like Shea a lot. Loved his Grantland writing, actually. Love his newsletter. Really loved The Hip-Hop Yearbook. But take a look at these…and tell me that 90-95% of them should be in existence. Brevity is a useful thing. This leads into the next issue(s).

4. Bill Simmons whiffed on getting almost all of his best writers to move to the new site from Grantland.

5. Because of this, nearly every genre is covered by the writing version of a running back committee with far less talented/creative individuals.

Just as an example, let’s look through FiveThirtyEight (another disappointing entity) and their list of Grantland’s best works. There’s 16 different articles on this list, and I’m withholding comment about how brutally perfunctory a list it is – where’s Bill Barnwell, Wesley Morris, Zach Lowe, Holly Anderson, Rembert Browne, etc.? Anyway, here’s the important thing: excluding Simmons, two of the 11 other writers represented followed Simmons to The Ringer. Again, this is excluding the four outstanding writers I mentioned above, plus Jonathan Abrams (who did appear on Bill’s podcast), Mark Harris, Sean McIndoe, Alex Pappademas, and more. Simmons has brought some of the good writers/editors from Grantland to his new project, but they either work in specialized areas (Juliet Litman and podcasts), are underutilized (Jordan Conn, two longform pieces in three months plus a pair of interviews), or are overutilized (Joe House, golf podcast?!?).

I continue to get the feeling that Simmons is hiring respected names without wondering if they’ll fit into his site’s plan, if there is one. Every tech writer hired is well-respected in that field, but they continue to push out stuff that’s borderline unreadable and generally not actual news. They bragged about not posting an iPhone 7 review, for God’s sakes! It also doesn’t help that Simmons lost his normal eye for talent (Dan Fierman) to MTV and that Simmons is now unusually focused on a fledgling TV show on HBO. Speaking of which…

6. The show sucks.

It really, really does. And it might get cancelled soon. Then what?

7. Simmons is proud of the variety of podcasts without doing much in the way of quality control.

Simmons’s best hire will end up being Chris Vernon, who hosted a radio show in Memphis for several years. Want to know why Vernon somehow seems comfortable on audio and not nervously laughing at anything even mildly funny his guest or co-host mentions? He has actual audio experience. No one else on the staff does. Simmons is comfortable because he’s done this for nearly a decade. Grantland‘s podcasts worked because they paired people together as sort of a buddy system so they could learn together and become better co-hosts – Robert Mays and Bill Barnwell, Juliet Litman and Chris Ryan, Chris Ryan and Andy Greenwald, Jalen Rose and David Jacoby, etc. It worked very well there. Pairing people like former Grantland intern Tate Frazier – whose voice is as exciting as the quiet drone of a ceiling fan – and Mallory Rubin, someone legitimately exciting, is very awkward and stilted all around. And every single Ringer podcast is like that. Some people aren’t meant to record audio, Bill.

8. Finally: it’s just Grantland-lite. Even the site design is bad in comparison.

Look at how gigantic and gross the banner ad is. The premier story on the site, as chosen by editors (I presume) takes up another 600 pixels or so. You can literally fill your entire screen with one story, a banner ad, and a list of other places on the site you can go to also be disappointed. Grantland’s site design is perfectly clean and actually featured multiple stories below that on a normal landing page, as the screencap above is cut off. It’s even frustrating to look at!

---

For Bill Simmons to fix his site and make it not suck (or at least return it to his original Grantland ideal), there are three remaining options:

1. Nix the entire tech section.
2. Prepare for cancellation of the television show and begin easing the TV writers into the Ringer site.
3. Stop the (Insert Vague Subject Here) Weeks. No one enjoys them. Just write.

What’s most disappointing about Bill Simmons’s vision and The Ringer as a whole is this: I see exactly zero of these happening. It’s one thing to disappoint; it’s another to promise a grander vision and present a product that looks like the Dollar General version of what you’ve done before.

Finding the upset: week 6

We’re moving in the right direction again. Last week’s picks of Oklahoma State over Texas, Indiana over Michigan State, and California over Utah gave us a record of 3-1, while we missed on Vandy over Florida (which was still a very close call). My system did miss on which 80-89% game would be the upset, but as it did say, there would be one. It just happened to be South Alabama over San Diego State. It also suggested an upset in the Top 11-15, which came through when North Carolina defeated then-#12 Florida State.

What I’m saying is that this system clearly appears to be pretty good at what it’s doing, and I’m now 8-5 straight up through four weeks (normal Vegas underdog win rate: 21%, metrics win rate: 61.5%). I think there are also ways to make it better, though we’ll have to explore those in future weeks.

This week has the Top 25 included:

With regards to the number of games in each setting, that refers to the games that aren’t Top 25 vs. Top 25 games. This week, we’ve got eleven total. When removing the four Top 25 vs. Top 25 games, that gives us a composite of 8.147333 expected wins by the Top 25 favorites, or an average of 2.853 upset losses.

As usual, the win expectancy categories, with a new one this time out:

• 30-39% (0-0 last week, 0-0 this season): #21 Colorado (at USC) (32.7%)
• 40-49% (0-2 last week, 2-3 this season): #18 Florida (vs. LSU) (45.3%)
• 60-69% (1-1 last week, 7-2 this season): #5 Washington (at Oregon) (68%)
• 70-79% (1-1 last week, 6-1 this season): #24 Utah (vs. Arizona) (74%), #20 Oklahoma (vs. Texas) (74.1%), #15 Stanford (vs. Washington State) (74.2%)
• 80-89% (4-1 last week, 7-2 this season): #6 Houston (at Navy) (81.4%), #19 Boise State (at New Mexico) (83.9%), #3 Clemson (at Boston College) (89.6%)
• 90-99% (3-0 last week, 31-1 overall): #4 Michigan (at Rutgers) (95.4%), #2 Ohio State (vs. Indiana) (96.2%)

Now, the likelihoods for each category:

• 30-39%: 0.327 of 1 (32.7% chance of 1-0)
• 40-49%: 0.453 of 1 (45.3% chance of 1-0)
• 60-69%: 0.68 of 1 (68% chance of 1-0)
• 70-79%: 2.223 of 3 (40.7% chance of 3-0)
• 80-89%: 2.549 of 3 (61.2% chance of 3-0)
• 90-99%: 1.916 of 2 (91.8% chance of 2-0)

There’s only one really obvious pick this week, and it’s USC over #21 Colorado. I don’t pass up on the rare chance that I get to pick an underdog that’s actually a 67.3% favorite. The second one is also pretty simple, as it’s a combination of playing the numbers and playing an obvious situation: LSU over #18 Florida. For the third, I’m unsure of what direction to go.

We can go with one or the other: 0.508 expected losses from the Top 5 teams this week or, alternately, 0.777 losses from the 70-79% range teams. When in doubt, go with the group with more expected losses – sorry, Oregon, but I can’t pick you over #5 Washington right now. (Or in the near future.) The issue with the 70-79% teams: all three are extremely close in probability, with each having a near-equal chance at a loss. Let’s examine them situationally.

#24 Utah just wrapped up an extremely disappointing loss to a very fun Cal team, but they get a pretty mediocre Arizona team at home that has two wins this year: Grambling State and Hawaii, neither in impressive fashion. #20 Oklahoma defeated a TCU team that’s starting to look pretty bad. #15 Stanford, who just got the living snot beaten out of them by Washington, is about to play a suddenly-hot Washington State team that’s dropped 107 points in their last two games.

I don’t really like any of these choices – I’d personally prefer Oregon over any of them, because the offensive talent is still there – but I’ll roll with Washington State over #15 Stanford. It’s tough to really be overjoyed about this pick from a subjective view, but here’s a couple good reasons on my end: Washington State is easily the best team of the four underdogs (computer composite rankings: Washington State #41, Oregon #61, Texas #66, Arizona #72). Stanford head coach David Shaw also has an unfortunate flaw on an otherwise strong college resume: after his first season (2011), Shaw and the Cardinal have lost at least one game each season as a 6.5 point favorite (13-6 straight up as a 5 to 11 point favorite, with an expected record of 14-5). Saturday’s spread? 7.5 points, down from an opening line of 12.5.

Now, let’s roll to the Top 25 vs. Top 25 matchups. Same as above, with only higher-ranked teams listed:

• 40-49% (0-1 last week, 2-3 overall): #10 Miami (FL) (vs. #23 Florida State) (47.9%)
• 50-59% (0-1 last week, 2-3 overall): #17 North Carolina (vs. #25 Virginia Tech) (58.7%)
• 60-69%: #8 Texas A&M (vs. #9 Tennessee) (63.4%)
• 80-89%: #1 Alabama (at #16 Arkansas) (84.2%)

And the categorical probabilities:

• Top 1-5: 0.842 of 1
• Top 6-15: 1.113 of 2 (30.4% chance of 2-0)
• Top 16-25: 0.587 of 1

The average number of Top 25 v. Top 25 upsets this week is 1.457, so the model leans more towards one upset than two. However, I’ll highlight a second one to look at as well for reasons further discussed. So, the automatic pick here for me is a team ranked from 6th to 15th. That’s either #8 Texas A&M losing to #9 Tennessee or #10 Miami (FL) losing to #23 Florida State. As much as I’d personally love to go with Tennessee over A&M, I fear it’s #23 Florida State over #10 Miami (FL), the much more likely event per the metrics.

A secondary pick would have to come outside of the top 5 and 6-10 if it were to happen. Let’s look at #25 Virginia Tech over #17 North Carolina as a strong possibility. For the record: the model has calculated an average of 5.28 expected losses by higher-ranked Top 25 teams in the last three weeks; there have been five. Adding this week to it would give a four-week expectation of 6.734, or around seven upsets. Let’s take a shot with two this week.

Last week’s picks were Oklahoma State over #22 Texas (W 49-31), California over #18 Utah (W 28-23), Indiana over #17 Michigan State (W 24-21), and Vanderbilt over #23 Florida (L 13-6) for a season-long record of 8-5. This week’s picks are USC over #21 Colorado, LSU over #18 Florida, and Washington State over #15 Stanford. The Top 25 vs. Top 25 upset picks are #23 Florida State over #10 Miami (FL) and #25 Virginia Tech over #17 North Carolina.

Finding the upset: week 5

All things fun have to end eventually. In this blog’s case, it ended with crashing through the floor and an 0-3 record on upsets last week despite two of our picks leading their games with 30 seconds to play. Football is beautiful, dumb, and frustrating. Let’s try to get back on track. Here’s this week’s spreadsheet…

I’ve included the Top 25 picks as a bonus because I want to get into those later, but the more I look at this, the more I think we could see upsets in pure numbers this week more so than anything truly shocking. I have extreme doubts that any of the top 10 teams playing an unranked opponent will lose this week, and I don’t think there’s even a feasible potential loss until you get to #15 Miami playing a somewhat difficult Georgia Tech team on the road. (More on that later.)

The Top 25 picks will help more as a guide to show what we were missing when we didn’t include them: without the T25 v. T25 games included, I wasn’t giving you the complete picture of how many Top 25 teams were likely to lose on a given weekend. That’s now fixed.

Anyway, the numbers you care most about: the models individually expect 3.85 upsets (S&P+), 3.71 (FPI), and 3.73 (Massey), creating a total number of around 3.76 anticipated losses. We can round that to the nearest whole number somewhat easily (four), so let’s find four upsets.

First up, your list of Top 25 teams versus unranked opponents, as ranked by win likelihood:

• 40-49% (1-0 last week, 2-1 through 3 weeks): #22 Texas (at Oklahoma State) (40.3%), #21 TCU (vs. Oklahoma) (49.2%)
• 50-59% (1-1 last week, 2-2 overall): #18 Utah (at California) (50.5%)
• 60-69% (3-0 last week, 6-1 overall): #17 Michigan State (at Indiana) (60.4%), #14 Miami (FL) (at Georgia Tech) (67.9%)
• 70-79% (2-0 last week, 5-0 overall): #12 Florida State (vs. North Carolina) (74.5%), #16 Ole Miss (vs. Memphis) (79.3%)
• 80-89% (2-0 last week, 3-1 overall): #23 Florida (at Vanderbilt) (80.2%), #13 Baylor (at Iowa State) (84.4%), #24 Boise State (vs. Utah State) (86.6%), #15 Nebraska (vs. Illinois) (87.3%), #19 San Diego State (at South Alabama) (87.4%)
• 90-99% (2-0 last week, 28-1 overall): #6 Houston (vs. UConn) (93.9%), #1 Alabama (vs. Kentucky) (98%), #2 Ohio State (vs. Rutgers) (98.6%)

Now the expected number of wins for each category:

• 40-49%: 0.895 of 2 (19.8% chance of 2-0)
• 50-59%: 0.505 of 1 (50.5% chance of 1-0)
• 60-69%: 1.283 of 2 (41% chance of 2-0)
• 70-79%: 1.538 of 2 (59.1% chance of 2-0)
• 80-89%: 4.259 of 5 (44.7% chance of 5-0)
• 90-99%: 2.905 of 3 (90.7% chance of 3-0)

Just like we’ve done each week so far, let’s bring in each category of the Top 25 (Top 1-5, 6-10, etc.) and check their likelihoods:

• Top 1-5: 1.965 of 2 (96.6% chance of 2-0)
• Top 6-10: 1.811 of 2 (81.9% chance of 2-0)
• Top 11-15: 3.141 of 4 (37.3% chance of 4-0)
• Top 16-20: 3.757 of 5 (20.7% chance of 5-0)
• Top 21-25: 2.56 of 4 (13.8% chance of 4-0)

This is a pretty easy curve to look at, and one that’s just very obvious. The likelihood of the category going undefeated decreases significantly past the top 10, and gets worse the further down you go. At the very least, we can expect about one upset each from the three lowest categories. This week’s three top upset picks are Oklahoma State over #22 Texas, California over #18 Utah, and Indiana over #17 Michigan State.

Reasoning for each: Oklahoma State could have cashed last week if Mike Gundy knew how to run a fourth-and-goal play call and/or didn’t fumble; Cal appears to be this year’s Team No One on Earth Enjoys Playing, and it seems pretty obvious that Michigan State is in rebuilding mode. More quantifiable, they all host the lowest chances of winning in their individual pods. I’d like to hunt down one in the Top 11-15 range, but I don’t like any of the situations the underdogs are in at all: Georgia Tech looked positively woeful offensively against Clemson last week and Miami’s had two weeks to prep for the triple option; Baylor losing to Iowa State seems unfathomable, no matter what season; Illinois is atrocious; UNC has easily the best odds of an upset, but this is a situationally poor game for them (emotional comeback victory over Pitt previous week and playing against a pissed-off Florida State team that can run like crazy).

To find a fourth upset, let’s look back at those win categories. Notice anything interesting? My eyes immediately jump to that 80-89% win range, which has the third-lowest likelihood of going undefeated. Pretty shocking, no? That means someone’s looking at a major haul of an upset. Again, we base this on likelihood, so you can rule the Top 16-20 out, which has no 80-89% range teams. The Top 11-15 has two in Baylor and Nebraska, but the likelihood of that group losing one of those plus the Miami game is just 4.5%.

Let’s roll back to the Top 21-25, which already fell alarmingly close to two upset losses. There’s two options here: a Florida team that just gave up 38 in a row to Tennessee or a Boise State team that is fine, but nothing beautiful. Neither one really excites me as an upset pick, either: Florida looked pretty great for one half against the Vols, and Boise’s already beaten two teams significantly superior to Utah State. The best pick I’ve got is situational here: Florida just came off of a major rivalry loss to a team that hadn’t beaten them in nearly 12 years, is missing their starting quarterback, and is about to play a team whose defense has looked pretty good in three of four games. (Georgia Tech is nearly always forgivable.) So, in true “throwing your hands up” fashion, eff it: Vanderbilt over #23 Florida is the stretch pick by the metrics this week, as it’s the most likely event of the 80-89% group. There are two things I love about this game: the under on 40.5 (neither offense will top 21 by themselves) and situationally both ways. Florida finished the Tennessee game on the losing end of a 38-7 run, while Vanderbilt just beat a solid Western Kentucky team on the road by scoring a TD on the final play then stopping WKU’s two-point attempt in overtime. They’ll be riding high while Florida will be travelling to Nashville to play a game they likely have little interest in on a noon kickoff.

Last week’s picks went 0-3, making us 5-4 on the year. This week’s upset picks for Top 25 teams playing unranked opponents: Oklahoma State over #22 Texas, California over #18 Utah, Indiana over #17 Michigan State, and Vanderbilt over #23 Florida as the stretch pick of the week per the metrics.

Now, let’s look at the Top 25 games. By the numbers, we’re looking at a composite total of 1.52 upsets from these four games – basically a coin flip between one or two upsets, and we met the expected total of 1.81 (2) last week. If we’re being honest, some won’t really even be upsets, but I digress.

Let’s run some win percentages:

• 20-29% (0-1 last week, 0-1 overall): #8 Wisconsin (at #4 Michigan) (23.3%), #25 Georgia (v. #11 Tennessee) (27.8%)
• 40-49% (1-1 last week, 2-2 overall): #3 Louisville (at #5 Clemson) (48%), #10 Washington (v. #7 Stanford) (48.6%)
• 50-59% (1-1 last week, 2-2 overall): #7 Stanford (at #10 Washington) (51.4%), #5 Clemson (v. #3 Louisville) (52%)
• 70-79% (1-0 last week, 1-0 overall): #11 Tennessee (at #25 Georgia) (72.2%), #4 Michigan (v. #8 Wisconsin) (76.7%)

Again, the Top 25 categories for each involved higher-ranked team:

• Top 1-5: 0.753 expected losses
• Top 6-10: 0.486 expected losses
• Top 11-15: 0.278 expected losses

So, if you’re a betting man, the obvious immediate look is to either #3 Louisville or #4 Michigan losing. Let’s take the much more likely event, both subjectively and objectively: #5 Clemson over #3 Louisville. If you need a second one, it seems that the Top 6-10 is a lot more likely than Tennessee losing to what smells distinctly like an 8-4 Georgia team. If there’s going to be a second upset this week, it’s likely #10 Washington over #7 Stanford. But, again, if I’m looking to bet, I lean away from looking at a second upset. We’ll see. By the metrics, our picks to meet the 1.52 (rounded up to 2) “upsets” are #10 Washington over #7 Stanford and #5 Clemson over #3 Louisville. If you’re just picking one, take #5 Clemson over #3 Louisville.

Finding the upset: week 4

We’re rolling large! After a Thursday night stumble in which Houston soundly defeated my Cincinnati upset pick (though Cincinnati did lead entering the fourth quarter), all three upsets singled out by my model rang true on Saturday. The two weeks I’ve ran out the Excel-based metrics model that draws from Bill Connelly’s S&P+, ESPN’s FPI, and the Massey Ratings, the model has nailed all five upsets that have happened between a Top 25 opponent and an unranked opponent.  It even sold me on North Dakota State over formerly #13 Iowa, which is a gigantic deal! Thanks to Mark Helfrich being a poor head coach, we came through on Nebraska over #22 Oregon. Even California and Sonny Dykes beat #11 Texas.

As a note, the normal Las Vegas underdog win rate is around 21%. I’m at 83% after two weeks and six games (5-1). This is meaningless because small sample sizes are terribly meaningless, but if I can end up breaking 60% by season’s end, we might have something really special here.

This week, the models expect anywhere from 3.55 losses in 12 games (S&P+) to 3.04 (Massey). Collectively, they average an expectation of 3.3 upsets this weekend. We finished one below expectation last weekend, so just as it was last week, we’ll include three upset picks and a fourth stretch pick as sort of a regression to the mean.

For the first time, I’m including my metrics below, with Top 25 v. Top 25 teams included and highlighted:

And without, which is much more important to this article.

Spreadsheet legend for the numbers on the top right: the top number is the total number of expected wins from that group; the bottom number is the percentage likelihood that the noted group goes undefeated.

As usual, the groups by win percentage are listed below:

• 40-49% (1-1 last week): #24 Utah (vs. USC) (47.3%)
• 50-59% (1-1 last week): #7 Stanford (at UCLA) (55.6%), #18 LSU (at Auburn) (58.8%)
• 60-69% (3-0 last week): #16 Baylor (vs. Oklahoma State) (62.1%), #20 Nebraska (at Northwestern) (65.2%), #13 Florida State (at South Florida) (66.2%)
• 70-79% (1-0 last week): #9 Washington (at Arizona) (75.8%), #5 Clemson (at Georgia Tech) (79.9%)
• 80-89% (0-1 last week): #4 Michigan (vs. Penn State) (82.4%), #3 Louisville (at Marshall) (85.5%)
• 90-99% (8-0 last week): #6 Houston (at Texas State) (91.6%), #1 Alabama (vs. Kent State) (99.4%)

The expected win totals for each of these groups:

• 40-49%: 0.473 of 1
• 50-59%: 1.144 of 2 (32.7% chance of 2-0)
• 60-69%: 1.935 of 3 (26.8% chance of 3-0)
• 70-79%: 1.557 of 2 (60.6% chance of 2-0)
• 80-89%: 1.679 of 2 (70.5% chance of 2-0)
• 90-99%: 1.91 of 2 (91.1% chance of 2-0)

Now, let’s note that the 60-69% group went 3-0 last week. That’s unlikely to happen again, not least because Georgia needed a fourth and 10 touchdown from a true freshman QB with a minute left to beat Missouri, but also because of these probabilities. The odds of this group going 3-0 last week were just 25.3%; the chances of the group moving to 6-0 over two weeks would be 6.7%.

Anyway, let’s make some picks. Based on the above probabilities in the Excel spreadsheet, here’s the expected number of wins for each fifth of the Top 25, broken down further:

• Top 1-5: 3.47 of 4 (56% chance of 4-0)
• Top 6-10: 2.23 of 3 (38.6% chance of 3-0)
• Top 11-15: 0.662 of 1 (66.2% chance of 1-0)
• Top 16-20: 1.86 of 3 (23.8% chance of 3-0)
• Top 21-25: 0.473 of 1 (47.3% chance of 1-0)

If we’re looking for three upsets, we’re going to hit the following categories, as they’re the most likely to find upsets: Top 6-10, Top 16-20, and Top 21-25. Adjusting this for the above win probability categories and composite ratings that show us one upset win for each of the three lowest groups in win percentage (40-49%, 50-59%, and 60-69%), we get the following most likely upsets: USC over #24 Utah, UCLA over #7 Stanford, and Oklahoma State over #16 Baylor.

Let’s rationalize these as best as we can, starting with the pick I hate the most. USC has looked about as excited to start the Clay Helton regime as I am when my friend group decides on Regional Tex-Mex Restaurant of the Week. This isn’t quite as much about USC’s strengths as it is about Utah’s very prominent weaknesses. The Utes currently rank 98th in offensive S&P+ and have yet to drop more than 34 points on any opponent despite drawing Southern Utah and San Jose State in two of their first three games. New QB Troy Williams has already thrown four interceptions, and he’ll be playing a USC secondary that is far and away the least-awful part of a very disappointing defense thus far. If Utah wins this, it’ll be because they topped 200 yards on the ground, not because of anything Williams does or doesn’t do. I don’t see at all how Utah can score enough to put USC away, so I’m taking Sam Darnold and company on the road here.

UCLA, much like their crosstown partners, are also waiting to post an impressive performance of some kind, though at least UCLA forced overtime with their opening week SEC sparring partner. UCLA’s offensive points per game is lagging far behind their statistical expected output, which usually means a breakout is nigh. Why not against a Stanford defense performing far ahead of what the stats expect? Regression to the mean is very real in sports, and I think this might be a great test case for it. Critical to UCLA’s success: force Ryan Burns to pass and you will win. He’s yet to attempt more than 18 passes in either of Stanford’s first two games, and Christian McCaffrey has yet to really look like Christian McCaffrey.

Oklahoma State is a great buy-low. They came out flat in a weird noon home game against a secretly good Central Michigan team, won a close one anyway, then lost on a play that shouldn’t have happened. They play a Baylor defense that allowed 36.3 points per game to top 25 offenses in 2015 and likely isn’t any better. Oklahoma State rebounded from the CMU loss by dumping 45 points on a Pat Narduzzi defense against Pittsburgh. I think people are selling this OSU team in the wrong spot because of the wacky loss; win this and they have a great October schedule to get back in the national conversation.

Now, if we’re anticipating a fourth upset to make up for the lack of a fourth last weekend, we need to go big again. We found gold in the 80-89% range with North Dakota State last weekend, but it was situationally perfect: Iowa at home with a mediocre offense against a great coach and team that was underrated because of being in the FCS. I would prefer to fade that this week, as expecting the rough equivalent of a 1 in 7 event to happen two weeks in a row is fool’s gold. That’ll rule out #4 Michigan and #3 Marshall. We’ve already picked out the three groups most likely to find an upset with one upset from each level of confidence, so that takes away #18 LSU, #20 Nebraska, and #13 Florida State. It is incredibly unlikely that a 90+% team loses, so remove #6 Houston and #1 Alabama. That leaves us with 70-79% and two games: #9 Washington at Arizona and #5 Clemson at Georgia Tech.

Let’s go back and look at that AP Top 25 breakdown: we’ve already bought our upsets from Top 6-10, Top 16-20, and Top 21-25. By taking out Florida State from consideration, we’re ruling out Top 11-15. That leaves the Top 5. Which one of these teams is the most likely team to lose from the Top 5, though it’s far from a certainty? Clemson. Let’s throw a dart and have some fun with a game in a weird and uncomfortable situation for the favorite: Georgia Tech over #5 Clemson on a Thursday night in Atlanta.

Here’s my rationale: Clemson technically has all of four days to run and prepare for a triple option offense. They finally looked like the Clemson everyone anticipated seeing this year on Saturday, but it was against woeful South Carolina State. Georgia Tech, meanwhile, dropped 38 points and 516 yards on a top 20 Vanderbilt defense. Paul Johnson is quietly pretty successful against Clemson and Swinney, having more wins against the Tigers than anyone not named Florida State. ACC favorites in Thursday night games are also just 7-6 over the last three years. I’m not totally positive this one happens, but after upsets coming from the 90-99% and 80-89% groups the last two weeks with one per week, it looks like it’s 70-79%’s turn this time out. Sorry, Clemson. (Alternately, if this one misses: sorry, Washington.) As the metrics only see three upsets, I don’t advise this one, but it’s worth a roll of the dice towards the season total. We’ll use this as a subjective test vs. the metrics, and I will still count it to my total.

To recap, our picks last week: Houston over #6 Cincinnati (miss), California over #11 Texas (make), North Dakota State over #13 Iowa (make), and Nebraska over #22 Oregon (make). Through two weeks, we’re 5-1.

This week: Utah over #24 USC, UCLA over #7 Stanford, Oklahoma State over #16 Baylor, and Georgia Tech over #5 Clemson (stretch pick, subjective choice).

Finding the upset: week 3

Last week’s first edition of upset spotting worked so well that the streets demand a new episode. Well, not exactly – I just really like doing these. New week, new probabilities!

My two upset picks went 2-0 this Saturday both on ridiculous finishes, with Central Michigan winning on a play that shouldn’t have happened and Arkansas using a wide receiver pass to get to overtime to beat TCU. I don’t know that we’ll need that much luck again, but we’ll see.

Anyway: out of the Top 25 teams, eight are playing each other, which leaves 17 teams that will be playing unranked teams. ESPN’s FPI system ran back at me with the following: based on their win probabilities, 13.19 of these 17 teams will defeat their (likely) underdog opponents. Bill Connelly’s S&P+ sees almost the exact same, coming in at 13.38. This week, we’re going to use the Massey Ratings as well, which I use for lots of prediction work. They roll in with a prediction of 13.17 games won out of 17, which makes every system in near-harmonious agreement. Just like last week, we’ll merge the metrics for a group of ~3.75 teams that will lose this weekend. The odds of all 17 teams going 17-0 this weekend are hilariously miniscule: 0.8%. Upsets are coming.

The trick: finding what 3.75 (alternately, four) of these teams will lose.

This week’s categories of victory likelihood, per the merged metrics:

• 40-49%: #17 Texas A&M (at Auburn) (43.2%), #22 Oregon (at Nebraska) (45.6%)
• 50-59%: #11 Texas (at California) (51.7%), #25 Miami (at Appalachian State) (59.7%)
• 60-69%: #6 Houston (at Cincinnati) (61%), #16 Georgia (at Missouri) (64.3%), #7 Stanford (vs. USC) (64.5%)
• 70-79%: #20 LSU (vs. Mississippi State) (76.1%)
• 80-89%: #13 Iowa (vs. North Dakota State) (86.6%)
• 90-99%: #4 Michigan (vs. Colorado) (91.3%), #21 Baylor (at Rice) (95.2%), #9 Wisconsin (vs. Georgia State) (95.8%), #15 Tennessee (vs. Ohio) (95.9%), #24 Arkansas (vs. Texas State) (97.1%), #8 Washington (vs. Portland State) (98.5%), #23 Florida (vs. North Texas) (98.6%), #5 Clemson (vs. South Carolina State) (99.9%)

Now, the number of games that each of these groups would win on average:

• 40-49%: 0.888 of 2 (19.7% of winning both)
• 50-59%: 1.114 of 2 (30.1% of winning both)
• 60-69%: 1.898 of 3 (25.3% of going 3-0)
• 70-79%: 0.761 of 1 (76.1% of winning)
• 80-89%: 0.866 of 1 (86.6% of winning)
• 90-99%: 7.723 of 8 (75.2% of going 8-0)

Based on this, we can expect to see at least three upsets, one from each of the three least confident percentage lines. Based on the odds, I’m not sure we will see a ground-shaking upset by the numbers, but anything can happen. To achieve a fourth, we’ll have to reach a bit from one of these groups, but we can get to that later.

Our initial upset picks for week 3 are as follows: Nebraska over #22 Oregon from the 40-49% group, California over #11 Texas from 50-59%, and Cincinnati over #6 Houston from 60-69%.

My reasoning for each, starting with Nebraska: Oregon’s defensive efficiency against two very poor offenses and teams (2-9 in FCS UC Davis and a Virginia team that just lost to an FCS team) has come in at an unadjusted 84th in America per ESPN. Tommy Armstrong is unbelievably frustrating, but the odds on him having a strong game here are very high. Another key stat: under head coach Mike Helfrich, the Ducks are 2-5 (0-3 OOC) in games where they have between a 40-60% chance to win, suggesting a lack of consistent ability to win close games. The projections expect almost exactly one loss on average from teams ranked 21-25 this week, and Oregon has far and away the worst odds of winning their game from that group.

California is a less certain pick for me, as my friend Harrison knows very well of my considerable disdain for one Sonny Dykes. I think he’s a smart offensive coach who has zero idea how to win at a Pac-12 program, simply because he has zero idea what he wants defensively. I’ll set that aside for this week alone, because their 48.3% odds of winning are both alarming and eye-opening. I do like QB Davis Webb a lot, but this comes from projection systems being very skeptical of one home win against a poor Notre Dame offense meaning the true comeback of Texas football. Also, Charlie Strong is 2-4 in his career in September road games against Power 5 opponents. For teams ranked 11-20, the projection systems anticipate a 4.178-1.822 record, and Texas’s odds were the worst of the 50-59% range.

Cincinnati over Houston will appear to be a very bizarre pick to just about everyone, but the projection systems were unusually strong in pushing back against a Top 10 team: neither FPI nor S&P+ give Houston greater than a 58% chance to win, and Massey’s 72% even comes with the caveat that they expect it to be a one-touchdown defeat for a pesky Cincinnati team. But: the collected projections anticipate around 0.89 losses from the six Top 10 teams playing this week against unranked opponents, and it was either Cincinnati on a Thursday night in a weird spot for Houston or expecting USC to win a big game after two very disappointing performances. My comfort in this pick comes from a combination of this being a terrible situation for Houston (four days rest, a near seven-hour game this past Saturday, Ward Jr. battling shoulder issues, and a long travel week from Houston to Cincinnati) and of Cincinnati’s surprisingly strong offense, anchored by QB Hayden Moore (7.7 YPA, 147.5 QBR) and a solid rushing game. The fun part: Cincinnati allowed six yards per play to Purdue, but has picked off opposing quarterbacks seven times through two games. I look forward to seeing if Tommy Tuberville has the capability left to force Greg Ward, Jr. into tough situations. I would also prefer to take Missouri over Georgia here, but this is a test for these systems and for my own subjective biases.

There’s the three big picks I’m making this week. Now, let’s get a little weird for the fourth.

Percentage-wise, the Top 11-20 is much more likely to lose twice than once this week, so our fourth upset is most likely to come out of this group. Here’s where it gets wild: the chances of all ten 70-99.9% teams going undefeated are just 49.6% (still high, but more likely that one loses than doesn’t), the number of games this group wins on average is 9.35 of 10, and the number of upsets expected from the 40-69% range is 3.1, which means we’ve exhausted our supply there. To find the fourth .65 of an upset, we’re going big. Keeping it in the Top 11-20 range, this means eliminating #11 Texas (who we’ve already picked against), #16 Georgia, and #17 Texas A&M. Congrats! That leaves us with the following three schools to pick from for a giant loss: #13 Iowa (vs. North Dakota State), #15 Tennessee (vs. Ohio), and #20 LSU (vs. Mississippi State).

Tennessee has by far the best odds of beating their opponent, and I feel confident and unbiased in ruling them out from an earth-shattering loss. Teams that were underdogs to Kansas do not beat good teams, sorry. So, let’s pick from Iowa or LSU for the stretch upset. There are solid cases for both this weekend: Iowa was nowhere near as good as the polls suggested last season (finished 47th on Football Outsiders) and still isn’t very trusted by advanced metrics this year, polling at 35th on FO and 25th on ESPN’s FPI. LSU had an unbelievably bad loss to Wisconsin to open up the season and still doesn’t have a quarterback. They actually both have similar odds on ESPN to defeat their lower-tier opponent: LSU at 86.4% for a Mississippi State team that lost to South Alabama, Iowa at 90.9% for FCS powerhouse North Dakota State, who would be ranked above 46 FBS teams right now. I expect neither of these upsets to happen, if I’m being honest.

But: when my hand is forced and I have to pick one to happen, why not North Dakota State over #13 Iowa? Do you really trust Iowa? Do you really think they’re going to do the undefeated thing again? Have you watched Kirk Ferentz football at any point before 2015? (A reminder that an 11-2 Iowa team came within a blocked field goal of losing to FCS Northern Iowa in 2009.) Also, NDSU has won five in a row against FBS opponents, is coming off of their fifth straight FCS national championship, and Iowa has not gone undefeated in out of conference play in the Kirk Ferentz era. Again, I don’t anticipate this happens – I have much more faith in #17 Texas A&M losing to Auburn – but this is worth a flyer.

As a reminder, last week’s picks were Central Michigan over #22 Oklahoma State and Arkansas over #15 TCU, both of which came through. At 2-0 on the season, we’ll roll the dice with Nebraska over #22 Oregon, North Dakota State over #13 Iowa (stretch pick), California over #11 Texas, and Cincinnati over #6 Houston.

Finding the upset: Week 2

The following is true, at least as true as something can be based on a computer model: Bill Connelly’s S&P+ picks and rankings say that on an average weekend with these matchups, 2.571 of the 23 Top 25 teams playing this weekend would lose. That would normally seem pretty fair if the following weren’t also true: 22 of the 23 are playing in games with win probabilities of 66% or higher (TCU at 51% over Arkansas). Per similar Football Performance Index (FPI) metrics on ESPN, the same games spit out odds of 1.313 teams playing this weekend losing. As you can tell, S&P+ is typically more aggressive, FPI more conservative. Merging the two, 1.942 of these 23 teams will lose this weekend, which seems fair. It’s time to figure out what two games these systems could be referring to and to figure out how there’s just a 24.2% chance the Top 25 teams playing this week go 23-0.

The odds of all five of the current Top 5 teams in America winning their leisurely waltz games is a solid 84.7%, so we won’t even touch those. Ditto that of the AP Top 10, of which nine teams are playing. There’s a 79.5% chance that those teams go 9-0, which is very high. As much as Wisconsin being ranked #10 makes me laugh, it won’t be Akron that unseats them. (Though perhaps one or two of these teams will struggle with their inferior opponents.) But 11-20 has a high chance of bearing at least one loss: the odds of the nine teams in this group playing this weekend going 9-0 are just 41.9%. Still high, but it’s more likely than not that at least one of these teams will lose. TCU has by far the lowest chance of victory in the top ten this week: 69.4% on FPI, 51% on S&P. Removing them from the equation shows that the other 8 teams have a 60.3% chance of going 8-0.

By the numbers and thanks to a spectacularly awful defensive performance last week – 41 points allowed to FCS South Dakota State – I’ll bite the bullet and roll with Arkansas over #15 TCU in one of the rarest cases of Septembert happening.

If we’re looking for a second upset, there’s no obvious contender. Louisville is around a 75% favorite over Syracuse and Tennessee about 77%, but the odds of both teams winning are around 57.8%. I tend to favor FPI’s idea that we’re looking at just one upset this weekend. If there’s going to be a second upset, it will be one well off the beaten path and a true shocker, which is why I’m willing to roll some rather extreme dice here: Central Michigan over #22 Oklahoma State.

Why this game? Simple probabilities. I was inclined to take one of #13 Louisville, #16 Iowa, or #17 Tennessee as a loser, but the three combine for around a 65.6% chance of going 3-0 per the more accurate FPI metric (for Week 1’s results, anyway; S&P+ says 34% and a 2-1 record). Alternately, I have no desire to pick Syracuse or Iowa State to do anything of importance and Virginia Tech is in the second game of a new coaching regime following an okay-ish performance against Liberty. There are no 70-79% or <60% win categories for teams this week, which made the unfortunate TCU at 69.4% this week’s semi-lock. But amongst the 90% and above crowd, there’s an astounding 19 teams with odds higher than 90.1% to win their game. The odds of this group going 19-0 stand at just 53%, which means they’re more likely than not going to do so, but that gives them a total of 18.39 games won out of 19. Basically: there’s a 47% chance that a 90% or higher favorite loses on Saturday. 90% or higher favorites on the S&P+ model won more than 96% of their games last season (6 losses total, and one of those was Baylor playing with a wide receiver at quarterback), which sounds about right, but with the same 162 game sample and the average of a 95% win probability (around what it was in this category in 2015), this suggests that they overachieved a bit: the average result would be 8 losses in 162 games, and if 2016 regresses to the mean, you’ll see more shock upsets. (You might also want to note that both 60-69% and 70-79% favorites overachieved their expected value in the regular season by an astounding +10 and +12 games after going -12 (a 22 game swing!) and +1 in 2014, respectively. Upsets are coming, and these are very volatile.)

Somewhat useful tangent having been achieved, I’m rolling with Central Michigan for the following reasons: they have the second-highest shot out of the 90% club to beat their favored opponent (7.9% on ESPN, 14.9% on S&P+), they’re unusually high in rankings for a MAC school (63rd/2nd in MAC on FPI), and in OOC games where Oklahoma State was favored by 20 or more in the last three years (they are currently favored by 20.5 in this game), they’re 2-4 ATS with one of those spread losses coming just last year at Central Michigan (23 point favorite, won 24-13). Also, for some less quantifiable factors: Cooper Rush (CMU QB) is likely the best Group of Five quarterback (2015: 66.3% completion rate, 25 TD/11 INT, 7.87 YPA, 13th in ESPN’s Expected Points Added metric), Central Michigan very nearly Did It last year (led in third quarter, trailed by four entering fourth), and Oklahoma State was unable to top four yards per carry against Southeastern Louisiana, which bodes…not well. (Garbage time factors in, of course, but Oklahoma State also posted 5.44 yards per play against an FCS team.)

We’ll see if these come to fruition, of course. The point of this is to exploit supposedly easy schedules and probabilities, because nothing is 100% guaranteed to happen in college football. You’ll see at least one upset this weekend and quite possibly another. Keep your apps ready.