VMI
Expected points
YardLine's expected-points model values every snap by what a drive from that spot is worth on average, so a drive can be judged against where it started. How it works →
Drives vs. expectation, 2018
Points over expected per drive, from where each drive started.
Per drive against an average FBS opponent, neutral field.
Points over expected per trip inside the opponent's 40.
Average expected points where drives started.
Every snap has an expected-points value: what a drive from that spot, down and distance is worth on average. A drive over expected beat its starting spot; defense counts what VMI allowed, so lower is better. The adjusted number fits every team against the opponents it actually played, so it leans toward average early in a season.
Game by game
Bars above the line: better than expected for VMI.
Fourth downs
YardLine's fourth-down model values going for it, kicking and punting on every real fourth down, so each call can be checked against the numbers. How it works →
Fourth-down calls vs. the numbers, 2018
Calls that matched the option with the best modeled win probability.
Summed over every call that differed, in points of win probability.
Went for it when going beat every other option by a point or more.
Each real fourth down is valued three ways (go for it, kick, punt) from the score, clock and field position before the snap, using league-wide odds rather than VMI's own. A call that differs gives up the gap to the best option; gaps under one point are close calls, and the 0 clear-cut ones account for 0.00 of the 0.00 expected wins. The model sees the situation, not the injury report, so a call it disagrees with can still be right.
Costliest calls against the numbers
- at ODUQ2 14:07 4th & 2 at VMI 37, down 28Punted; the numbers favored going for it (close call)−0.1 pts
- at ODUQ3 11:56 4th & 6 at ODU 49, down 56Went for it; the numbers favored punting (close call)<0.1 pts
- at TOLQ1 10:38 4th & 2 at VMI 33, down 7Punted; the numbers favored going for it (close call)<0.1 pts
Every call this season (23)
- Q1 15:004th & 14 at VMI 21, tied · PuntedAgreed
- Q1 10:384th & 12 at VMI 11, tied · PuntedAgreed
- Q1 10:384th & 2 at VMI 33, down 7 · PuntedClose call
- Q1 7:254th & 10 at VMI 40, down 14 · PuntedAgreed
- Q2 13:094th & 5 at VMI 30, down 28 · PuntedAgreed
- Q2 8:424th & 9 at TOL 10, down 28 · Kicked a field goalAgreed
- Q2 4:544th & 6 at VMI 23, down 32 · PuntedAgreed
- Q2 2:444th & 14 at VMI 24, down 39 · PuntedAgreed
- Q3 13:134th & 6 at VMI 34, down 39 · PuntedAgreed
- Q3 11:124th & 15 at VMI 20, down 46 · PuntedAgreed
- Q3 2:274th & 5 at VMI 30, down 60 · PuntedAgreed
- Q4 14:134th & 6 at TOL 32, down 60 · Went for itAgreed
- Q1 11:264th & 17 at VMI 46, down 7 · PuntedAgreed
- Q1 3:024th & 3 at VMI 26, down 21 · Went for itAgreed
- Q2 14:074th & 2 at VMI 37, down 28 · PuntedClose call
- Q2 9:184th & 3 at ODU 42, down 42 · Went for itAgreed
- Q2 5:154th & 7 at VMI 28, down 49 · PuntedAgreed
- Q2 3:304th & 10 at VMI 27, down 49 · PuntedAgreed
- Q3 11:564th & 6 at ODU 49, down 56 · Went for itClose call
- Q3 4:424th & 4 at ODU 4, down 56 · Went for itAgreed
- Q4 13:194th & 1 at VMI 22, down 49 · Went for itAgreed
- Q4 10:054th & 10 at VMI 25, down 63 · PuntedAgreed
- Q4 10:054th & 7 at VMI 39, down 56 · PuntedAgreed
After touchdowns: kick or go for two?
Tries that matched the option with the better modeled win probability.
Summed over every try that differed, in points of win probability.
Went for two when it beat kicking by a point or more.
Win probability
YardLine's model estimates each team's chance of winning before kickoff and again after every play, from the score, clock, down, distance and field position. How it works →
Record vs. expectation, 2018
Expected wins adds up VMI's chance of winning each game as the model saw it at kickoff - a 70% favorite counts 0.7.
What moved it, 2018
Every game starts at VMI's pregame chance; everything after that is the field. The model's own breakdown of each game's final play splits the on-field part by what drove it, and the parts add up exactly. How it works →
- Scoreboard−0.05
- Less the pregame edge, counted before kickoff+0.02
Game by game
- TOLL 3-66 at TOL0% → 0%Before kickoff −50 · On the field 0: Scoreboard −1
- ODUL 14-77 at ODU2% → 0%Before kickoff −48 · On the field −2: Scoreboard −4
Points of win probability, from VMI's side.
Season arc, 2018
Every completed game this season, in order - each tile is VMI's own win-probability curve.
Biggest swings of 2018
The largest single-play change in VMI's win probability this season - one snap that moved the outcome the most.
Stolen wins
VMI's own lowest win probability at any point in a game it went on to win, across the whole archive - the deepest holes it climbed out of.
Let it slip
VMI's own highest win probability at any point in a game it went on to lose, across the whole archive - the biggest leads it didn't hold.
Stock & rivalries
Stock
VMI's leaguewide Elo rating, tracked like a stock price - built from game results and margins, not the model's own win-probability predictions.
Series & rivalries
VMI's most-played opponents across the whole archive - open one for the full head-to-head history.