Expected points, explained

The idea in one sentence

Expected points is the average net score — your points minus the opponent's — that historically followed from a given situation, before the next play happens. Net is the important word. If you have the ball on your own five-yard line, the number is negative, because often enough the next score in the game belongs to the other team.

What the numbers look like

Rough values for first-and-ten, taken from decades of play-by-play data. First-and-ten on your own five is worth about −0.4 net points. At your own twenty, roughly 0.5. At midfield, about 2.0. At the opponent's twenty, around 4.0. First-and-goal at the two is near 5.5 — close to a touchdown, but not a touchdown, because the goal-line stand is real.

Notice how uneven the scale is. The first fifty yards of field position are worth about 1.5 points. The next thirty are worth two more. Field position compounds as you approach the end zone, which is why a punt that pins an opponent inside their own ten is worth so much more than one that lands at the twenty-five.

Down and distance

Field position sets the ceiling; down and distance decides how much of it you keep. At the same spot on the field, first-and-ten might be worth 2.0 points, second-and-eight about 1.7, third-and-six roughly 1.1, and fourth-and-six barely 0.3 — because on fourth down the drive is usually over one way or another.

Distance interacts with down in a way that pure yardage does not capture. Third-and-one and third-and-nine are the same down but nearly different sports: conversion rates run around 70 percent versus 35 percent. This is why a two-yard run on first down is a decent play and the same two yards on third-and-eight is close to a punt.

Expected points added

The number becomes an evaluation tool the moment you take differences. Expected points added, or EPA, is the value of the situation after a play minus its value before. A seven-yard gain on first-and-ten at midfield takes you from about 2.0 to about 2.4 — plus 0.4. A sack on third down might cost 1.2. An interception swings the sign of the whole number, because possession changes hands: your positive becomes their positive from the new spot.

EPA is a far better measure of a play than yards, because it prices context. Four yards on third-and-three is a good play; four yards on third-and-eight is a bad one. Yardage cannot tell them apart. EPA per play has become the standard efficiency metric for exactly this reason.

How EP drives fourth-down decisions

Take fourth-and-two at the opponent's forty. A field goal is out of range. A punt nets maybe twenty-five yards of field position, worth roughly 0.4 points of swing. Going for it converts around 55 percent of the time; success gives you first-and-ten near the thirty-five, worth about 3.0, while failure hands the opponent the ball at the forty, worth about −1.5 to you. The weighted value is 0.55 × 3.0 + 0.45 × −1.5, which is about 0.98 — comfortably better than the punt. This arithmetic, repeated across every fourth down, is why coaching has changed.

Where the model has to be adjusted

Late-game score effects

Expected points is score-agnostic by construction. Down twelve with two minutes left, points are not what you need — you need points quickly, and a slow scoring drive that drains the clock can be worth less than a failed fast one. This is exactly where win probability replaces EP as the right lens.

Team quality

Baseline EP tables are league averages. An elite offence is worth more at every spot on the field than a poor one, sometimes by half a point or more in the red zone. Applying a flat table to a mismatch understates the favourite.

Level of play

College numbers differ from NFL numbers. Scoring rates are higher, defences are more variable, and field-position value curves are steeper. Borrowing one table for the other quietly biases everything downstream.

Using EP when you are pricing a live game

Convert to points, then to probability. If a team is 0.9 expected points better off after a drive, and the game's remaining variance is roughly ten points of standard deviation, that swing is worth something like three to four percentage points of win probability — more late, less early, because there is less time left for randomness to undo it.

The practical use is spotting overreactions. A live market that moves twelve cents on a play worth 0.4 expected points has moved too far, and a market that barely twitches on a red-zone turnover has not moved far enough. EP is what tells you which is which.