CFB · SCORECARD · SIMS · COLLEGE-FOOTBALL · ESPN-FPI · WEEK-8
The Sims Had Their Worst Week of the Season and Still Beat ESPN's FPI
By MVPeav · @MVPeav ·
37-19. Week 8 was my sims' worst week of the season, and I'm putting the number at the top because a scorecard that hides the ugly week is a brochure.
What keeps me from sulking is FPI. ESPN's number went 34-22 on the same 56 games. My sims beat it by three wins on a weekend where everybody's math got lit on fire.
BYU, Oregon and JMU all lost as big favorites inside the same 48 hours. No system saw that coming. The question is whether mine was built to survive it, and the answer is mostly yes.
The Ugly Record and the Better News Under It
66.1% is the low. The peak was 86.0% back in week 3, and since then it's been 69.6, 69.0, 69.5 and now 66.1. Four straight weeks under 70 is a trend, and I'm treating it like one.
The better news is who I was losing to. On the same 56 games, FPI went 34-22. On Brier, which is the average miss on the win chances where lower is better, my sims came in at 0.2337 against FPI's 0.2592.
Third straight week beating FPI on picking winners, too. FPI took weeks 2 and 5, and it hasn't won one since.
The score ranges held up on a weekend built to break them. Actual margins landed inside my middle-80% range 83.9% of the time, right where they should, and inside the middle-50% range 46.4% of the time.
Total points were the one place the week really showed. My average miss on the combined score was 15.4, against 13.0 for the season. When BYU and UCF play double overtime, the final score goes somewhere no median is going to find it.
Where the Sims Were Dead On
Five games where my sims and FPI picked opposite winners. My sims got 4 of them.
The one I'd put on the fridge is Tennessee at South Carolina. FPI took Tennessee. My sims had South Carolina at 59.0%, South Carolina won 27-24, and Tennessee's 24 was the exact median.
Then the Alabama game, because of course it's the Alabama game. The median sim said Texas A&M 20, Alabama 31. The final was Texas A&M 20, Alabama 31. Alabama's 31 was right, the total of 51 was right, the margin was right. 10,000 sims did the work, and I'm still going to enjoy it as a Bama fan.
Rutgers at Northwestern was the quiet one. Total points hit the median of 56 on the nose, and Aidan Chiles threw for 219 yards against a median of 219. Nobody is writing a song about Rutgers-Northwestern, but a sim that guesses a quarterback's yardage to the number is doing something right.
BYU, Oregon and JMU All Went Down at Once
Six of my favorites at 70% or better lost in week 8, and I'm not hiding any of them. The three that mattered are in the title. The other three were Troy (78.5%) to UL Monroe, Iowa (76.8%) at Minnesota, and NC State (74.4%) at Stanford.
JMU was the biggest number at 79.8%, and FPI had them even higher at 84.5%, so nobody was alone on it. It's also the one that surprised me most. App State won 34-31 in overtime. Colton Phares picked off a pass with about 6 minutes left to set up the go-ahead score, JMU tied it with 3 seconds on the clock, then missed a 47-yarder in OT.
The number underneath is the run game. JMU came in averaging 223.7 rushing yards a game and App State held it to 92 on 30 carries. A team that runs for 224 a week doesn't lose that game in my sims very often. A team that runs for 92 does, and that's the team JMU was on Wednesday.
BYU is the one with the biggest repercussions for the playoff. My sims had the No. 2 team at 79.2% at UCF with a median of 28-16. UCF won 39-31 in double overtime, the program's first win ever over a top-5 team. The miss hurts my scorecard; the loss moves the bracket.
Oregon is the one I can at least explain. 79.0% in my sims, 77.0% in FPI, median Oregon 34, Illinois 21. Oregon outgained Illinois 426 to 314 and lost 21-10, because Dylan Raiola fumbled twice, threw no touchdowns, and the Ducks went 4-for-15 on third down.
Illinois started Michael Clayton II for the first time in his career, got 11-of-18 for 84 yards and 2 TDs out of him, and scored 21 unanswered after going down 10-0. Oregon didn't score after halftime. It was Oregon's first Big Ten road loss since joining the league. You can outgain a team that badly and still lose, it just takes two fumbles and a third-down day like that one.
Malachi Toney Broke My Sims
The team misses had explanations. The player misses had Malachi Toney.
Miami's Malachi Toney caught 16 passes for 194 yards and 3 touchdowns. My sims had his medians at 6 catches, 72 yards and 0 scores. The 16 catches showed up in about 1 of every 1,000 sims. I ran that game 10,000 times and his Saturday lived in a corner nobody visits.
Josh Hoover went the other way. Indiana beat Michigan 13-6 and Hoover threw for 82 yards on 8 completions, against medians of 227 and 20. Any version of that game where Hoover completes 20 passes for 227 is the version my sims kept drawing, and Indiana won without it.
Auburn's Jackson Harris had 178 receiving yards in a game LSU won 38-17, against a median of 48. Arizona beat Iowa State by 36 when the median margin was 4. Those are the tails doing what tails do, and there's no setting I'd change to catch a 16-catch game.
Are the 70 Percent Favorites Really 70 Percent
Calibration is the test I care about most. If my sims call a team 75%, that team should win about 75% of the time. Mostly it does. Then there's the one bucket.
Favorites my sims put between 70% and 80% have played 99 games this season. They were supposed to win 75.0% of them. They've won 68.7%. BYU, Oregon and JMU were all sitting in that exact bucket on Saturday. The leak was there before week 8, and week 8 poured through it.
The tiers above it are fine, and if anything a little too modest:
- 80%-90% favorites: 60 games, expected 84.3%, won 88.3%
- 90%+ favorites: 55 games, expected 95.7%, won 94.5%
So the big favorites are a touch stronger than my sims say and the mid favorites are a touch weaker. If you wanted one fix, it's shaving a few points off anything in the 70s.
FPI's problem is at the top. Its 90%+ favorites have won 90.5% of 63 games against an expected 94.9%, and its 80%-90% favorites won 80.2% against 85.2%. My sims leak in one bucket. FPI leaks in two, and they're the two with the sure things in them.
Should You Still Trust the Sims This Week
Yes, with an asterisk.
Season to date through week 8, my sims are 288-105, which is 73.3%, with a Brier of 0.1858. FPI on the same 393 games is 286-107 with a Brier of 0.1888. Two wins and a hair of Brier over nearly 400 games is a lead, and I'm counting it.
FPI does have the better average margin miss, 11.5 points to my 12.0. FPI's scores land a little closer. My sims are better at saying who wins. Those are two different jobs and I can live with losing the one by half a point.
The head-to-head on disagreements is the sobering number. In the 44 games this season where my sims and FPI called opposite winners, my sims are right in 23. Barely better than half, and week 8's 4-of-5 helped.
The ranges are what I trust most. For the season, actual margins land inside the middle-80% range 81.9% of the time and inside the middle-50% range 53.9%. The range of outcomes is built right, even in the weeks the winner call is wrong.
So use them this week. Read the 80s and 90s straight, read the ranges straight, and when you see a favorite in the 70s, remember that's where BYU, Oregon and JMU were living on Saturday.