EPA in Football: What Every Play Is Worth

EPA, short for expected points added, measures how much a single play helps or hurts a drive. Every situation on a football field, from first-and-10 at your own 25 to first-and-goal at the 5, has an expected-points value built from years of real drives. A play's EPA is the value after the snap minus the value before it.

That's why EPA reads a game better than yards do. A 4-yard run that converts third-and-3 keeps a drive alive, while a 4-yard run on third-and-10 usually sets up a punt. Yards per carry scores both runs as 4.0, and EPA pulls them far apart. You can look up any team's or quarterback's EPA per play on rbsdm.com, which gets its play-by-play from nflfastR.

Expected points give every spot on the field a value

nflfastR's expected points model starts from the situation before the snap and asks what the next score of the game is likely to be and which team will get it. A touchdown counts as 7 and a field goal as 3, and when the other team is the likelier one to score next, the value drops toward zero or below it. Here's what the model said about some common spots, averaged over 2025 regular-season plays with more than two minutes left in the half:

Pinned at your own 10, an offense is worth almost nothing, because a punt from deep in its own end often hands the other team the next score. FTN describes the same curve in its own work: each yard is worth more as a team nears either goal line.

Calculating EPA takes one subtraction

Take the expected points after the play and subtract the expected points before it. Say it's first-and-10 at your own 25, worth 1.2. A 15-yard completion puts you at first-and-10 at your own 40, worth 2.0, so that play earned about +0.8. An incompletion on the same snap leaves second-and-10 at the 25, worth about 0.7, so it cost roughly half a point.

Scoring plays use the points themselves. nflfastR counts a touchdown as 7 and a field goal as 3, so a touchdown from first-and-goal at the 5 adds about 7 − 5.8 = +1.2. Add up every play for a team's total EPA, then divide by the number of plays for EPA per play, the version most people quote.

The same math explains a quirk in who gets credit. A 50-yard catch that sets up first-and-goal at the 2 earns a big chunk of EPA. The 2-yard touchdown run after it adds only about +0.7, since the offense was already expected to score about 6.3 points from there. FTN's DVOA explainer uses this exact case to argue about credit: an offense with first-and-goal at the 2 scores a touchdown five times out of six, so the runner who punches it in didn't do much of the work.

Only five offenses cleared 0.1 per play in 2025

New England's offense led the league at 0.155 EPA per play, and Las Vegas finished last at −0.192, per rbsdm.com. The average team landed just above zero. Only four other offenses got past 0.1, with the Rams closest behind at 0.149.

New England's 0.155 was about 0.14 better than an average offense, which comes to about 8.5 points over a game of 60 snaps. On defense, lower is better. Seattle allowed −0.113 per play, the best mark in the league, so opposing offenses lost value on a typical snap against it.

Success rate counts the plays that helped

Success rate is the share of plays that did the offense some good. On rbsdm.com a play counts as a success when its EPA is above zero, the definition in the nflfastR data dictionary. The older version grades yards against the sticks. FTN's running back success rate wants 40% of the needed yards on first down, 60% on second and all of them on third or fourth, and the success bar under DVOA asks for 45% on first down.

The rules can split on the same play. A 4-yard run on first-and-10 passes the 40% bar and misses the 45% one. By EPA it usually fails: in nflfastR's 2025 play-by-play, first-and-10 plays that gained 4 or 5 yards came out positive only 35% of the time.

Success rate and EPA per play can split at the top, too. Here's how four 2025 offenses looked on rbsdm.com:

The Rams stayed on schedule more than anyone. New England got to the top of the EPA list with big plays: its passing game produced 0.308 EPA per dropback, best in the league, while its running game ranked 25th at −0.089 per rush. Explosive plays lift EPA far more than they lift success rate, which is why the Patriots could lead one list and sit fourth on the other.

Red-zone touchdown rate is the noisy part

Red-zone efficiency is the share of trips inside the opponent's 20 that end in touchdowns. rbsdm.com lists it on its Luck page next to fumble recoveries, and the 2025 numbers show why. Philadelphia scored touchdowns on 70.5% of its 44 red-zone trips, the best rate in the league, with an offense that ranked 17th in EPA per play. New Orleans was last at 44.4%, and the league as a whole finished near 57%.

EPA already accounts for the red zone. First-and-goal at the 5 is worth about 5.8 expected points, so a drive that stalls there and kicks a field goal gives back about 2.8 points over that series, and a touchdown adds 1.2. An offense's EPA per play also reflects how often it gets that deep, and the Rams got there 76 times in 2025, more than any other team. When you size up a team, trust how often it reaches the red zone before you trust its touchdown rate once it's there.

For player picks, the red zone changes who matters. The back who gets the carries inside the 10 isn't always the one who leads the team in rushing, and a fade in the end zone goes to a different kind of receiver than a crosser at midfield. Reading those roles week to week is covered in how to read an NFL player's week.

DVOA adjusts for situation and opponent

DVOA stands for Defense-adjusted Value Over Average. Aaron Schatz created it at Football Outsiders, and FTN has been its home since August 2023. Like EPA, it grades every play by situation. FTN's own example: five yards on third-and-4 are worth more than five yards on first-and-10, and much more than five yards on third-and-12.

DVOA gets there with different math. It scores each play in what FTN calls success points, compares that with what an average offense does in the same situation, then adjusts for the quality of the opponent. The result is a percentage where zero means league average. A team at +10% was 10% better than average, and because the number measures scoring, a good defense shows up negative. In 2022, Kansas City's offense led the league at 25.2% and San Francisco's defense led at −14.1%, per FTN's explainer.

Early in a season, read DVOA with care. FTN's team DVOA page notes that opponent adjustments start only after Week 4. For its own in-season game picks, FTN uses DAVE, which mixes DVOA with its preseason projections, and keeps using it until teams have played 13 games.

EPA per play is the one to pull game by game on rbsdm.com, and it ties straight to points. DVOA does the opponent adjustment for you, which helps most once a few weeks of schedule have piled up. When the two disagree about a team, check its schedule first. For the quarterback version of the same idea, see ANY/A.

How EPA changes a Team Pick'em pick

Say your GAGE Team Pick'em slate asks whether an underdog wins, and YES shows a 30% chance. Lock YES and it's worth +7.0 if they win and costs 3.0 if they lose. The underdog ranks in the top 10 in EPA per play and success rate but has dropped a few close games, and the favorite's scoring leans on a red-zone touchdown rate like Philadelphia's 70.5%. EPA catches that kind of gap, and it's a reason to think 30% is low. If you're not sure what a 30% chance is telling you, start with how to read a favorite and an underdog.

The full table is on how GAGE scores a pick, and GAGE is on the App Store when you want your own reads scored.