An Advanced Stats Primer: Success Rate, Expected Points and Efficiency

By 4Football Staff · November 5, 2026 · 10 min read

Success rate, expected points added and efficiency metrics explained without the jargon: what each one measures, how to read it, and where analytics can mislead you.

Two offenses each gain 4.2 yards on a run. One of those runs came on first-and-10 and left the team with second-and-6. The other came on third-and-3 and moved the chains. The yardage is identical. The value to the offense is not even close.

Traditional statistics treat them the same. Advanced statistics exist to tell them apart. If you have heard announcers say "EPA" or "success rate" and nodded along, this primer is for you. It explains the main ideas in plain language, shows how to read each one, and, just as important, explains what they cannot do.

None of this requires math beyond multiplication. The concepts are simple. What makes the field complicated is that different analysts build their numbers differently, so the details vary from one site to another.

Why box score yards are not enough

Total yards and yards per carry count every gain equally. But a football play is part of a sequence with a purpose: get a first down, score, or stop the other team from doing so. Three things make a yard worth more or less than another:

• Down and distance. Four yards on third-and-3 is a win. Four yards on third-and-10 is a failure.
• Field position. Ten yards from your own 20 and ten yards from the opponent's 20 are not the same event. The second puts you in scoring range.
• Situation. The score, the clock and the quality of the defense all matter.

The measures below try to capture some or all of those factors. Most start from the same idea: evaluate each play by what it did for the offense's chance of scoring or winning, not just how far the ball moved.

Success rate

Success rate is the simplest advanced stat. Each play is labeled a success or a failure, and the success rate is the percentage of successes.

One widely used definition is:

• On first down, a play succeeds if it gains at least 40 percent of the yards needed for a first down (4 yards on first-and-10).
• On second down, it succeeds if it gains at least 60 percent of the yards needed.
• On third or fourth down, it succeeds only if it gets the first down or scores.

Other analysts use slightly different thresholds, and college versions often set the first-down bar higher. Before you compare success rates from two sources, check that they use the same rules.

A small example

Say an offense runs ten plays in a made-up game, drawn from several different drives.

Five of ten plays succeeded, so the success rate is 50 percent. Notice that play 2, a two-yard gain, was a failure, while play 3, a four-yard gain, was a success. Same-ish yardage, opposite meaning.

What it tells you

Success rate measures consistency. An offense with a high rate stays ahead of the chains and rarely faces long downs. It is a less flashy number than yards per play because it ignores how big the successful plays were. That is useful. A team that relies on a few long gains and many failures can have a decent yards-per-play figure and a poor success rate, and that style is harder to sustain.

A good pairing is success rate with explosive play rate, the share of plays that gain a big chunk of yardage. Definitions of "explosive" differ, often something like a run of 10 or more yards and a pass of 20 or more. Consistency plus big plays describes a very good offense.

Expected points

Expected points is the foundation of most modern efficiency analysis.

The idea is this. Look at every time in history a team faced a given situation, say first-and-10 at the opponent's 35-yard line, and measure what happened next: how many points did the team score on that drive, or what did the next score of the half turn out to be? Average the result. That average is the expected points (EP) of the situation.

Some consequences of how EP behaves:

• The closer to the end zone, the higher the EP.
• Deep in your own territory, EP can be negative, because the opponent is more likely to score next than you are.
• Better down and distance gives higher EP: first-and-10 beats third-and-10 from the same spot.

Expected points added

Now apply it to one play. Before the snap, the situation has some EP. After the play, the situation has a different EP. The difference is expected points added, or EPA.

EPA = EP after the play − EP before the play

Here is a hypothetical. Suppose a model values first-and-10 at the offense's own 25 at 0.6 points. After a 12-yard gain, it is first-and-10 at the 37, which the model values at 1.0. The play's EPA is +0.4. If instead the play had lost three yards, leaving second-and-13 at the 22, and the model valued that at 0.1, the EPA would be −0.5. (These values are invented to show the arithmetic. They are not real model outputs.)

Special cases

EPA deals naturally with plays that don't gain yards but change the situation:

• A touchdown is worth the points scored (plus the extra point's expected value) minus the EP before the play, so a long touchdown gets a large EPA and a one-yard touchdown from the goal line gets a small one.
• An interception or fumble sends the ball to the other team, and the EPA drops by the amount you lost plus the points the opponent now expects. A turnover deep in your own territory is far more costly than a turnover near the opponent's end zone.
• A punt is also scored, from the change in field position and possession.

Reading EPA

The most common form is EPA per play. Positive means the offense is gaining expected points; negative means it is losing them. Averaged over a season, a team's EPA per play is a strong summary of its offensive efficiency. Over the league as a whole, analysts have consistently found that passing plays produce more expected points per play, on average, than running plays. That finding is one reason offenses throw more now than they once did, and it does not mean any given run is a bad idea.

Variants to know:

• Dropback EPA per play counts only plays where the quarterback dropped back to pass, including sacks and scrambles. It is a more complete view of a passer than passer rating, because it includes sacks and ties each throw to the situation.
• Rush EPA per play does the same for designed runs.
• Defensive EPA per play is the same measure, taken from the defense's side.

Win probability

A related model estimates the chance that each team will win from any point in a game, using score, time left, field position, down, distance, timeouts and a measure of team strength. Win probability added (WPA) is the change on one play.

Win probability is the right tool for late-game decisions, because the value of points changes with the score. A touchdown when you trail by 20 early barely moves the needle, but a first down while leading by four with two minutes left may change it a lot. EPA ignores the score. WPA depends heavily on it, and that is also its weakness: a team's best plays in a blowout win barely register.

Efficiency metrics that need less machinery

Not all advanced numbers require a model of the whole game. Many are simple arithmetic on the standard record.

Another useful one, points per drive, matters because football is organized around possessions. A team can post a high yards-per-play figure and still score little if its drives end with turnovers or stalls at midfield.

Opponent-adjusted measures

One game against a strong defense and one against a weak defense are not the same test. Several systems try to adjust. The best known is DVOA, which stands for defense-adjusted value over average. It compares every play with the league average for that down, distance and opponent, and reports a team's results as a percentage better or worse than average. Other sites publish similar opponent-adjusted ratings.

The method is complicated, and each publisher uses its own version. What they share is the aim: a number that says how a team performed given who it played. When you see one, read it as a rating produced by a particular model, not an official statistic.

Tracking-data metrics

Since player-tracking chips and camera systems arrived, the league and its data partners have published statistics that were impossible to calculate by hand: how fast a passer takes to throw, how often a receiver separates from a defender, how much space a ball carrier had at the handoff.

• Completion percentage over expected compares an actual completion rate with the rate expected from the difficulty of the throws.
• Rush yards over expected compares a runner's yards with the yards that similar carries typically produce, which helps separate the back from the blocking.
• Pressure rate counts how often a defense disrupted a passer, though what counts as a pressure is partly a judgment, so providers differ.

These measures are new compared with the traditional record, and each depends on the provider's method. They are best viewed as informed estimates.

Pitfalls: how analytics goes wrong

Small samples

A single game contains well over 100 plays between both teams. A single player might see a few dozen snaps in an important role. Per-play averages swing wildly in samples that small. Wait for several games before drawing a conclusion, and be careful with splits (such as "third down in the red zone"), which cut the sample down further.

Luck and regression

Some outcomes are less repeatable than others. A fumble on the ground is recovered by one team or the other with something close to even odds, no matter how good either team is. Turnovers in general tend to move back toward average from one season to the next. When a team's results are driven by turnover luck rather than efficiency, expect them to fade. See our piece on regression to the mean for how that works.

Garbage time

Plays in the final minutes of a lopsided game are played under different incentives. The leading team runs the ball, the trailing team throws and the defense gives up short gains. Many analysts filter out these plays, with the cutoffs set differently from one analyst to another.

Model differences

Different expected-points models give different values for the same situation, so one site's EPA will not exactly match another's. Look at the trend and the rank, not the third decimal place.

Missing context

An analytics number cannot see a missed assignment, a receiver who ran the wrong route or a wind gust. It averages across thousands of plays, so it describes tendencies, not intentions. Coaches combine such figures with film.

How to read a stat line like an analyst

1. Start with the question. Are you asking about passing, running, defense or the whole team?
2. Pick a measure that matches. Dropback EPA per play for a passer, success rate for consistency, points per drive for scoring.
3. Compare to the league, not to zero. A number is only high or low relative to others in the same season.
4. Check the sample. How many plays are behind it?
5. Look for agreement. If three measures say the same thing, it is probably real.
6. Add what the numbers cannot see, such as injuries, scheme and weather.

Teams use these ideas in real decisions, and the sport has moved visibly toward more passing and more aggressive fourth-down play as a result. Our article on how teams use analytics describes how that works in practice.

Where this connects to what you already know

If you use passer rating, you now know why analysts prefer play-based measures for judging quarterbacks: they include sacks, rushing and context. If you read a box score, you can tell which of its numbers are raw counts and which would benefit from adjustment. The team pages, league leaders and scores on 4Football show the standard statistics that these advanced measures build on.

The short version

• Advanced stats judge a play by what it did for the offense, not just by how many yards it gained.
• Success rate measures consistency: did each play keep the offense on schedule?
• Expected points values every situation; EPA is the change in that value on a play.
• Win probability matters for late-game choices because it depends on the score and the clock.
• Opponent-adjusted systems like DVOA estimate performance given the strength of the opposition.
• Small samples, turnover luck, garbage time and model differences all limit what a number can tell you.
• Use several measures together, compare them with the league and keep the film in mind.