When LSU beat Florida on Dec. 12, it all but ended the Gators' national championship hopes. They potentially could have worked their way back into the picture with an upset of Alabama, but didn't quite pull it off.
It was even more frustrating when you looked under the hood, and saw that Florida losing to LSU was almost impossible. The Gators outgained the Tigers by 191 yards and 3.3 yards per play. They generated a success rate of 49% to LSU's 39%. They finished nine drives inside the LSU 35-yard line, and LSU finished only five drives inside of the Florida 35.
I reference a measure called postgame win expectancy quite a bit. It takes all the key predictive stats that a given game produces (a lot of the stuff that gets fed into SP+), tosses them into the air and says, "With these stats, you could have expected to win this game X% of the time." For Florida, the number against LSU was 99.4%.
But 99.4% isn't 100%. LSU stuffed a fourth-and-goal attempt on the first drive of the game, Kyle Trask threw a pick at the LSU 32 (which Eli Ricks returned for a touchdown) and, minutes later, another pick at the LSU 6. And late in the game Marco Wilson was flagged for the most unlikely, ridiculous and downright hilarious unsportsmanlike conduct penalty you'll ever see, for throwing Kole Taylor's shoe downfield after a third-down stop. That set up a 57-yard field goal from Cade York, and when Evan McPherson couldn't answer with a 51-yarder at the final whistle, LSU had pulled off a 37-34 stunner -- maybe the best game of the year, and definitely the least likely result.
Lowest postgame win expectancy in a win, 2020
1. LSU 37, Florida 34 (Dec. 12): 0.6%
2. Texas 41, Oklahoma State 34 (Oct. 31): 2%
3. Louisiana 24, UAB 20 (Oct. 23): 4%
4. Indiana 36, Penn State 35 (Oct. 24): 5%
5. Georgia State 36, Troy 34 (Oct. 24): 6%
6. California 21, Oregon 17 (Dec. 5): 6%
7. Louisiana Tech 31, Southern Miss 30 (Sept. 19): 6%
8. Oklahoma State 20, Kansas State 18 (Nov. 7): 7%
9. Memphis 30, Houston 27 (Dec. 12): 8%
10. Louisiana Tech 37, UAB 34 (Oct. 31): 10%
The very next week, LSU got outgained by 0.9 yards per play by Ole Miss and again generated an inferior success rate. The Tigers were poor in the red zone, too. Postgame win expectancy: 27%. But while they defensed (intercepted or broke up) seven passes, which typically equates to about 1.5 interceptions, they held on to five picks. They also recovered the game's only fumble as Ole Miss was moving into position for the game-winning score. LSU 53, Ole Miss 48.
The win column suggests LSU finished the year strong and gave fans something to look forward to. (Freshman Kayshon Boutte caught 19 balls for 416 yards and four scores in the two wins.) The underlying stats, however, reveal a pair of games that would have produced an 0-2 record nearly three-quarters of the time. LSU actually fell from 51st to 62nd in SP+ in the past two weeks.
Winning can build morale and chemistry, and provide proof of concept for a coach or staff. But the stats you produce are more predictive when it comes to knowing how a team will perform moving forward.
Wins vs. second-order wins
Adding a team's postgame win expectancies together for a full season gives you what I call second-order wins. (I stole that term from baseball sabermetrics.) They hint at what your record was most likely to be, based on the stats each game produced. Because football is such an impossibly random sport -- especially when played by 18- to 22-year olds -- your record and second-order win total can differ by a good amount.
Biggest positive difference between win total and second-order wins, 2020
1. Army (+2.3: 9-2 record, 6.7 second-order wins)
2. Louisiana Tech (+2.3: 5-4, 2.7)
3. Ball State (+2.3: 6-1, 3.7)
4. LSU (+2.1: 5-5, 2.9)
5. Louisiana (+2.0: 9-1, 7.0)
6. Memphis (+2.0: 7-3, 5.0)
7. Stanford (+1.9: 4-2, 2.1)
8. UTSA (+1.8: 7-4, 5.2)
9. Indiana (+1.7: 6-1, 4.3)
10. Miami (+1.6: 8-2, 6.4)
All of these teams were rather fortunate this year, something that probably wouldn't surprise you if you watched Indiana's last-second stunner over Penn State or Army's win over Air Force this past Saturday (27% postgame win expectancy), in addition to LSU over Florida. But when you see that Louisiana Tech is a ghastly 112th in SP+ despite a winning record, or Indiana is only 24th despite the second-best record in the Big Ten, this hints at why.
Miami's presence on this list tells you just how much randomness this game can provide. The Hurricanes produced a second-order win total of 8.7 last season but went only 6-7; that minus-2.7 difference was the worst in FBS. Was Manny Diaz the worst game coach in history, or was the Hurricanes' fortune about to turn? Turns out it was the latter.
Biggest negative difference between win total and second-order wins, 2020
1. Louisville (-2.8: 4-7 record, 6.8 second-order wins)
2. Virginia Tech (-1.8: 5-6, 6.8)
3. Arkansas (-1.7: 3-7, 4.7)
4. Texas State (-1.7: 2-10, 3.7)
5. UAB (-1.6: 6-3, 7.6)
6. Wyoming (-1.6: 2-4, 3.6)
7. Mississippi State (-1.4: 3-7, 4.4)
8. Southern Miss (-1.4: 3.7, 4.4)
9. Troy (-1.3: 5-6, 6.3)
10. Georgia Tech (-1.3, 3-7, 4.3)
Louisville completely dominated in all four of its wins (each had a postgame win expectancy of 97% or higher) but managed to lose games with expectancies of 77% (Virginia), 60% (Boston College), 58% (Miami) and 49% (Notre Dame). Virginia Tech had a similar fate, dropping games at 77% (Wake Forest), 67% (Liberty) and 54% (Miami).
Keep this in mind when 2021's win totals are released next spring, and both the Cardinals' and Hokies' numbers are pretty low.
Misleading bowl matchups
This bowl season has provided us with some outright oddities this year, from a team having played just five games (Colorado) to, of course, SEC teams with miserable records (2-8 South Carolina, 3-7 Mississippi State and Arkansas).
Had South Carolina not withdrawn from the Gasparilla Bowl, the Gamecocks -- who really were as bad as their record (second-order win total: 2.4) -- would have played a UAB team that not only won Conference USA but also should have been closer to 8-1 than 6-3.
Some other bowls will feature even more misleading records.
• Cure Bowl (Dec. 26): 11-0 Coastal Carolina vs. 9-1 Liberty (9-2 vs. 8-2 on average)
• First Responder Bowl (Dec. 26): 9-1 Louisiana vs. 7-4 UTSA (7-3 vs. 5-6)
• Cheez-It Bowl (Dec. 29): 8-2 Miami vs. 7-3 Oklahoma State (6-4 vs. 6-4)
• Arizona Bowl (Dec. 31): 7-0 San Jose State vs. 6-1 Ball State (6-1 vs. 4-3)
• Liberty Bowl (Dec. 31): 9-2 Army vs. 5-4 West Virginia (7-4 vs. 6-3)
• Outback Bowl (Jan. 2): 6-1 Indiana vs. 4-5 Ole Miss (4-3 vs. 5-4)
See a trend? Dream seasons -- such as the ones experienced by Coastal Carolina, Liberty, Louisiana, San Jose State and Indiana -- are often accompanied by a decent amount of fool's gold. We inherently know this, of course; knowing the ride isn't going to last is part of what makes it exhilarating.
Do any coaches consistently defy the numbers?
Postgame win expectancy is one of my favorite numbers to reference during the season, because of the misleading wins it can so clearly highlight. But if you're consistently winning or losing games you shouldn't, that might not be all random. Some game coaches are better than others, right?
With college football's tiny sample size, it takes a while for consistent over- and underachievers to show up in the data. But they do exist.
Looking at the seasons of 2005 through 2019, six coaches (a) were head coaches for at least 10 seasons, and (b) overachieved by at least 0.7 wins per year on average:
1. Ken Niumatalolo, Navy (12 seasons, +1.11 wins per year)
2. Bill Snyder, Kansas State (11, +0.96)
3. Urban Meyer, Florida/Ohio State (13, +0.86)
4. Pat Fitzgerald, Northwestern (14, +0.77)
5. Mack Brown, Texas/North Carolina (10, +0.76)
6. Kyle Whittingham, Utah (15, +0.73)
Niumatalolo (+1.8) and Fitzgerald (+1.5) overachieved their stats this year, too. (That's right, Navy overachieved by going 3-7. Yikes.)
In smaller samples, others have shown overachievement potential. In his first five seasons at Houston and Texas, Tom Herman averaged +1.3 wins per year, stemming primarily from a 3.3-win overachievement at Houston in 2015 and +1.7 at Texas in 2018. Both teams failed to live up to the hype the next year. This year Texas was on the "slightly fortunate" list: 5.4 second-order wins, 6-3 record.
Another coach who has started out looking like an overachiever: Neal Brown. His last three Troy teams (2016-18) went a combined 31-8, overachieving their second-order win totals by a combined 4.6. He came to West Virginia and produced a 5-7 record in 2019 despite just 3.6 second-order wins. A good game coach who will consistently outperform his stats? Maybe not -- West Virginia was on the unlucky list this year: 6.3 second-order wins, 5-4 record.
As you might expect, the list of coaches who both underachieve consistently and produce a large sample of seasons isn't very long -- underachieving usually gets you fired. But here's a list of steady underachievers with at least five years as a head coach during the 2005-2019 time frame:
1. Mark Whipple, UMass (5 seasons, -1.40)
2. Ron Zook, Illinois (7, -1.04)
3. Kevin Wilson, Indiana (6, -0.91)
4. Mike Bobo, Colorado State (5, -0.87)
5. Paul Haynes, Kent State (5, -0.86)
Zook produced some solid seasons, going 9-4 in 2007 and 7-6 in both 2010 and 2011. The Fighting Illini haven't had a winning record since his departure. But he also left wins on the table.
Wilson, meanwhile, built a lovely infrastructure in Bloomington, Indiana. The Hoosiers had produced a top-75 SP+ ranking just once in nine seasons before he was hired in 2011. They did so in each of his last four seasons, but never topped 6-7. Indiana couldn't break the curse until going 8-5 in 2019 and, of course, 6-1 this fall.