Before a play even happens, a situation already has a worth in points. Expected points puts a number on it - and that number is the foundation under EPA, 4th-down math, and the 2-point decision.
By The NFL Analytics Editorial Team · Published June 15, 2026
Football is played in situations, not just yards. A 1st-and-10 at your own 5-yard line and a 1st-and-10 at the opponent's 5-yard line are the same down and distance, but they are worlds apart in value. Expected points (EP) is the number that captures that difference. It answers a single question: given the down, the distance to go, and the spot on the field, how many points will the next score in the game tend to favor the team with the ball - on average, accounting for the chance the other team scores next instead?
The key word is average. Expected points is not a prediction about one specific play or drive. It is the long-run average net points that follow from a given situation, measured across thousands of historical plays that started in roughly that same spot. A situation worth +2.0 expected points does not mean the offense will score two points; it means that, on average, the offense in that exact down-distance-field-position will out-score its opponent by two points before the next change of the game state.
The single most important picture in this whole topic is how expected points changes as you move down the field. Hold the down and distance fixed - say 1st-and-10 - and slide from your own goal line to the opponent's, and the EP value climbs steadily. Backed up against your own end zone, a 1st-and-10 is worth only a little, and in the deepest situations it can even go negative, because the most likely next score is a safety or a short-field opponent touchdown after a punt. Out near midfield it crosses into clearly positive territory, and inside the red zone it rises toward the value of a probable score.
The relationships that drive the curve are intuitive once you see them stated plainly:
Those three levers - field position, down, and distance - are the core inputs of every expected-points model. Better models add more context (time remaining, score, timeouts), but the down-distance-field-position skeleton is what makes EP a situational value rather than a flat per-yard estimate.
Hypothetical, rounded values for illustration only - they show the shape of the curve, not any published model's exact output.
| Situation | Roughly worth (illustrative) |
|---|---|
| 1st & 10, own 5 | slightly below zero to near zero |
| 1st & 10, own 35 | around +1 |
| 1st & 10, midfield | around +2 |
| 1st & 10, opp 20 | around +4 |
| 1st & goal, opp 5 | well above +4, toward a likely TD |
Expected points is not a formula you can write on a napkin the way you can with passer rating. It is an empirical model fit to historical play-by-play data. The construction follows a clear logic, even if the statistics underneath get involved.
Start with the idea of the "next score." For every play in a large historical sample, you look ahead and record what the next scoring event was and which team produced it: an offensive touchdown (worth roughly +7 with the extra point), a field goal (+3), a safety conceded (-2), a defensive touchdown by the team that was on defense (-7 from the offense's perspective), or no further score before halftime. Each historical play thus gets tagged with the net points that actually followed it.
EP(situation) = average net "next-score" points across all historical plays in that situation
Then you group or model plays by their situation. The simplest version just averages the next-score outcomes within buckets of down, distance, and yard line. Modern versions replace the crude buckets with a smooth statistical model - historically a regression, and now commonly a flexible machine-learning model - that maps the inputs (down, distance, yard line, and often time, score, and timeouts) onto an expected next-score value. The model smooths out the noise so that neighboring situations have sensibly related values instead of jumping around because one rare play happened in one bucket.
A few honest caveats come with any EP model:
This is the distinction that trips people up most, and it is genuinely simple once stated: EP is a level; EPA is a change. Expected points describes the value of a situation before a play. Expected points added measures how much a single play moved that value.
EPA = EP(situation after the play) − EP(situation before the play)
Walk a single play through it. Suppose an offense faces 1st-and-10 at midfield, a situation the model values at roughly +2.0 EP. The quarterback completes a pass to the opponent's 30-yard line, setting up a new 1st-and-10 worth roughly +3.2 EP. The play's EPA is +3.2 − +2.0 = +1.2. The offense added 1.2 expected points with that snap. A sack, a turnover, or a stuffed run would have lowered the after-play EP below the before-play EP, producing negative EPA.
One more nuance: because a touchdown is a terminal event, the EP after a scoring play is just the points scored (adjusted for the ensuing kickoff/field-position handoff), which is how the model assigns full credit to the plays that finish drives. The bookkeeping at the boundaries - scores, turnovers, end of half - is exactly where EP models earn their keep.
The reason expected points is more than an academic exercise is that it converts strategy questions into apples-to-apples comparisons. Once every situation has a points value, you can value the outcomes of a decision and weigh them by how likely each outcome is. Two of the most famous in-game calls fall straight out of this.
On 4th down the offense has three options, and EP lets you put each on the same scale. Going for it is worth the conversion probability times the EP of a fresh 1st down, plus the failure probability times the (negative) EP of handing the opponent the ball at that spot. A field goal is worth the make probability times three points, plus the miss probability times the opponent's EP from the resulting field position. A punt is worth the negative EP the opponent gets at their expected starting spot. Whichever option has the highest expected points - net of what you give the opponent - is the EP-optimal call.
This expected-points framework is exactly why the analytics community concluded teams punt too often on fourth-and-short: the EP of converting in plus territory frequently beats the EP of giving the ball away, even after accounting for the cost of a failed attempt. (Late in close games the decision shifts from raw points to win probability, which weights those same outcomes by how much they change the chance of winning rather than the expected points.)
The same machinery decides whether to go for two. The expected value of the two-point try is two points times the conversion rate; the expected value of the extra point is one point times the kick's make rate. Compare the two expected-point values and take the larger - the break-even two-point success rate is just half the PAT make rate. It is the identical "weigh each outcome by its points value and its probability" logic that drives the fourth-down call.
Reading about expected-points decisions is one thing; turning the dials is another. We built a transparent, browser-based 4th-down decision calculator on the NFL Calculators page that uses exactly the expected-points model described above. You plug in the yard line, the yards to go, your conversion and field-goal estimates, and the expected-points values for each outcome, and it recommends Go, Punt, or Field Goal based on which option carries the highest net expected points.
If you want to compute the foundational quantity yourself, public play-by-play data carries an expected-points value on every snap, so you can recover EPA as the play-to-play change directly from the data.
Expected points puts a single number on a football situation: the average net points the offense can expect from a given down, distance, and field position, based on what historically happened next. The value rises as the offense moves toward the opponent's goal line, on earlier downs, and with shorter distances to go. EP models are fit empirically to play-by-play data by averaging the "next score" outcomes within each situation. EP is a level and EPA is the change between two levels, which is why every EPA figure rests on an expected-points model underneath. And because EP values outcomes on one scale, it converts the biggest in-game calls - fourth-down go/punt/kick and the two-point try - into a clean comparison of expected points weighted by probability. Turn the dials in the 4th-down decision calculator to feel how the math works.
Want the code behind these metrics? Work through the 45-chapter NFL analytics tutorial.
Browse tutorials Free tools