Three efficiency metrics, one decision: what each measures, where they agree, and why they sometimes disagree.
By C. B. Zakarian · Published June 16, 2026
EPA, success rate, and DVOA are the three efficiency metrics you will see most often in modern NFL analysis. They are frequently treated as interchangeable, but they are not - each was built to answer a slightly different question. Used together they give a fuller picture than any one alone; used carelessly they can flatly contradict each other.
This guide defines each metric, explains what it is best at, shows how they can disagree, and makes the case for reading them as a set rather than picking a single favorite.
Every situation on the field - down, distance, and field position - has an expected points value: the average number of points the offense will eventually score on this drive, given thousands of historical plays from that same spot. EPA measures how much a single play changed that expectation.
EPA = Expected Points (after the play) - Expected Points (before the play)
A 30-yard completion that flips a team from its own 20 to the edge of field-goal range adds a lot of expected points. A sack that backs the offense up and burns a down subtracts them. Touchdowns and turnovers produce the biggest swings. Crucially, EPA is built from open play-by-play data - the nflfastR model is public, so anyone can compute it and inspect the assumptions.
EPA's strength is that it captures the full magnitude of a play. It naturally rewards explosiveness, because big plays move expected points the most. Summed over a season it becomes a total-value measure; averaged per play it becomes an efficiency rate.
Success rate is built directly on top of EPA, but it throws away the magnitude and keeps only the direction:
Success Rate = Share of plays with EPA > 0 (positive EPA)
A play is "successful" if it added expected points - that is, if it left the offense better off than before. Success rate is the fraction of plays that clear that bar. A 2-yard run on 3rd-and-1 and a 60-yard bomb both count as exactly one success; the difference in size is ignored on purpose.
This makes success rate a measure of consistency and floor. A team with a high success rate is staying on schedule, avoiding negative plays, and not constantly facing 3rd-and-long. It tells you how often an offense is doing its job, not how spectacularly.
DVOA (Defense-adjusted Value Over Average) also grades plays against a situational baseline, but it adds two things EPA does not include by default: an explicit opponent adjustment, and proprietary weights that emphasize successful plays. The result is reported as a percentage above or below league average.
DVOA ≈ Opponent-adjusted value over average, expressed as a % (0% = average)
DVOA's strength is fair comparison. Because it bakes in schedule strength, it is well suited to ranking teams who have faced very different opponents. The trade-off is transparency: unlike EPA, the exact weights are not public, so you cannot fully reproduce it. For the full breakdown, see our DVOA explainer.
| Metric | Best at measuring | Opponent-adjusted? | Open data? |
|---|---|---|---|
| EPA/play | Total value & explosiveness | Not by default | Yes (nflfastR) |
| Success rate | Consistency & floor | Not by default | Yes (derived from EPA) |
| DVOA | Opponent-adjusted efficiency | Yes, explicitly | No (proprietary weights) |
The most instructive case is when two metrics point in different directions. Consider two hypothetical offenses (numbers invented for illustration):
Hypothetical figures, chosen to make the point.
This is the whole reason to read them together. A single number can flatter a team that is actually volatile, or undersell a team that is quietly consistent.
You want total value, you care about explosiveness, or you need a transparent metric you can compute and audit yourself from public data.
You want to know how consistent or dependable a unit is, or you want a stabilizing sanity check on a small, noisy sample.
You want a single opponent-adjusted comparison across teams with very different schedules, and you are comfortable with a proprietary model.
EPA captures the magnitude of every play and rewards explosiveness; success rate strips that down to a yes/no to measure consistency and floor; DVOA adds an explicit opponent adjustment and reports efficiency as a percentage over average. They will sometimes disagree - a boom-or-bust offense can post strong EPA with a weak success rate - and those disagreements are exactly where the insight lives. The right move is to read them as a set: how much, how often, and against whom.
Want the code behind these metrics? Work through the 43-chapter NFL analytics tutorial.
Browse tutorials Free tools