Prediction
Nobody Had Rutgers: The Model's Big Ten Board for Week Two
Big Red Nation, the model had a very good week and a very instructive one, and those turned out to be the same week.
One thing before the numbers, because it matters. Everything below is the model’s output — a computer program, run Thursday morning, picking games off ratings and recruiting and returning production. These are not my picks. I do not overrule it, I do not tune it toward what I want to happen, and where I think it is wrong I say so in plain English rather than changing the number.
My own pick for Nebraska and Bowling Green lands Saturday morning, separately, with my name on it. If the two disagree, that is the interesting part, and you will be able to see both.
Week One: the model went 12-2
Straight up — just picking winners — it went 12-2. Season to date 16-4. On the games it labels strong, it is 12-1.
Against the spread it went 8-5-1. That is the harder test: not who wins, but whether the model beat the betting line. Eight times it landed on the right side of the number, five times it did not, and once the result finished exactly on the line — a push, which counts as neither a win nor a loss. Season to date that is 11-8-1.
Average score error: 8.65 points a team.
Those are good numbers. Now the part that matters more.
The two it missed, and who else missed them
Massachusetts 37, Rutgers 21. The model had Rutgers 45-7. That is its worst game of the season so far — and it is worth knowing what the professionals had.
The betting line was Rutgers by 29.5. Massachusetts won by 16. Every oddsmaker in Nevada was 45.5 points wrong on that football game. The model was wrong in exactly the same direction as the entire market, just further out on the same limb.
Nobody had Rutgers. Not us, not Vegas, not anyone taking money on it.
Michigan State 30, Toledo 20. The model took Toledo outright at 55.5% — a toss-up, flagged as a toss-up — and lost it. The line on that game was Michigan State by 10, and Michigan State won by exactly 10. The market nailed it to the point. That one the model lost on its own, and that is the honest way to say it.
How close is this thing to Vegas, really
Across all 14 games last week, average error on the final margin:
| Average margin error | |
|---|---|
| The model | 14.5 points |
| The betting market | 13.6 points |
Less than a point apart. And on Michigan against Western Michigan — the line said Michigan by 27.5, Michigan won by one — the model came in ten points closer to the truth than the market did.
That is the useful frame for everything below. This is not a machine that beats Vegas. It is a machine that lands in roughly the same neighbourhood as Vegas, and occasionally sees a game the market has mispriced. When the two disagree, that is worth reading. When they agree and are both wrong, that is football.
The pick I would not touch
San Diego State 24, UCLA 23. The line has UCLA by 12.5. That is a 13-point disagreement, the widest on the board — and the model is not even confident about it. 51%. A pure coin flip, the least conviction of any game it picked.
So the model is saying it has no idea, in a game where the market thinks it does. That is not a disagreement worth acting on — it is the model shrugging loudly. I would not go near it. It is printed here because printing only the picks that work is how you end up with a model nobody should trust.
Second-widest gap, and you will enjoy this: Rutgers again, by 8.7 more than the line, at Boston College. After last week, read that one for laughs.
The model on Nebraska: 38-10
Saturday, 6:00 p.m., Memorial Stadium, under the lights. It is the model’s most confident pick on the whole board at 97%, projecting Nebraska by 28.
The line says 29.5 — so for once the model is within a point and a half of Vegas, which almost never happens. Its oldest habit is coming in under the market, and last week against Ohio it came in eleven points light on a 49-21 game.
The rest of the board
Every line live from the book this morning.
| Game | Model | Line |
|---|---|---|
| Oregon over Oklahoma State | 38-10 | 23.5 |
| USC over Louisiana | 41-14 | 30.5 |
| Penn State over Temple | 35-17 | 23.5 |
| Washington over Utah State | 35-16 | 26.5 |
| Nebraska over Bowling Green | 38-10 | 29.5 |
| Iowa over Iowa State | 30-14 | 14.0 |
| Michigan State over Eastern Michigan | 31-20 | 17.5 |
| Maryland over UConn | 31-21 | 12.5 |
| Illinois over Duke | 28-23 | 6.0 |
| Minnesota over Mississippi State | 28-24 | 1.5 |
| Ohio State over Texas | 28-23 | Texas by 1.5 |
| Wake Forest over Purdue | 27-24 | Wake by 3.0 |
| Oklahoma over Michigan | 24-21 | Oklahoma by 5.5 |
| San Diego State over UCLA | 24-23 | UCLA by 12.5 |
| Rutgers over Boston College | 30-24 | BC by 3.5 |
Three Big Ten teams are picked to lose — Purdue, Michigan and UCLA. The market agrees on the first two and violently disagrees on the third.
Ohio State at Texas is the game. The model has Ohio State by five on the road; the line has Texas by a point and a half. A seven-point gap on the best matchup of the weekend — and unlike the UCLA pick, the model is at 63% rather than 51%, which means it is reading something rather than shrugging. That is the disagreement I would defend.
Worth knowing about this slate: 14 of the 30 teams on it return less than a quarter of last season’s passing production. Half this board is opening with a quarterback nobody has data on.
Husker Nation, six o’clock Saturday, under the lights. My Bowling Green pick lands that morning.