The most profitable football bettors have both a broad understanding of statistics and the wisdom to know which numbers matter most. Today, we’ll dive into the specifics of expected points added (EPA) per play, a pivotal part of any football bettor’s toolbox. You can use EPA to make sharper power ratings to fuel your projection-based betting models.
Note that projection-based betting isn’t the only way to be a profitable NFL bettor. Our Portfolio EV betting tools use a market-based methodology to do so. You can check out our +EV betting results here; the same approach works for football, basketball, baseball and more! If you want to know how we do it, read our guide to NFL betting strategy.
What are Expected Points Added (EPA)? NFL Betting Guide
What are Expected Points Added (EPA)?
So, what are expected points added (EPA), and what is EPA per play? Across sports, the analytics movement has developed “expected” statistics that calculate what an average team will do in a given situation, allowing analysts to identify which teams beat expectations and which do not.
Let’s go term-by-term here. Expected points are the value of a given possession by down and distance from the opponent’s end zone. A 1st and 10 from your 30-yard line will, on average, lead to fewer points than a 1st and 10 from your opponent’s 30-yard line.
Expected points added, then, are what the offense adds (or defense subtracts) to that possession. Move the ball from your 30-yard line to your opponent’s 30-yard line, and congratulations! You’ve added expected points. Take a sack or turn the ball over? You’ve lost expected points.
To get expected points added per play, simply divide the total number of expected points added a team generate during the game by the number of plays they ran. Operating on a per-play basis as opposed to a per-game one gives us a larger sample and helps normalize the data.
Next, we’ll dive into a case study argue why you should care about EPA and how you can use it to become a better bettor. But if you’re still confused about what these words mean — or you’re just more of a visual learner — check out the video I put together below:
Using EPA to Make Predictions: A Case Study
Let’s get into some proof of concept. Through Week 10 of last season, the top-10 teams in offensive EPA per play included the usual suspects, like the Kansas City Chiefs and Buffalo Bills, as well as the season’s surprises, like the Philadelphia Eagles, New York Giants, Seattle Seahawks and Jacksonville Jaguars.
Pause. Yes, I said the Jacksonville Jaguars. Although it’s easy to remember how good they looked down the stretch, let’s not forget that after Week 10, the Jaguars were 3-6 and third in the AFC South. They trailed the 6-3 Tennessee Titans and 4-5-1 Indianapolis Colts, leading only the 1-7-1 Houston Texans.
After Week 10, the Titans ranked 24th in EPA per play, ahead of the 30th-ranked Texans and 31st-ranked Colts. A bettor using EPA would have correctly identified the Jaguars as the best offense in the division long ahead of their strong run down the stretch. The Jaguars went 6-2 after Week 10. The rest of the AFC South went a combined 3-21. I wonder if anyone saw that coming?
Using EPA to Evaluate Offenses
You can use EPA per play to evaluate both offenses and individual quarterbacks. Obviously, the quality of a team’s quarterback is heavily linked to the quality of its offense — the Kansas City Chiefs would be a very different team if Chad Henne had to replace Patrick Mahomes — so it’s best to understand EPA with that in mind.
Some offenses can replace quarterbacks without skipping much of a beat. The San Francisco 49ers rotated between three different starting signal callers last year and almost made the Super Bowl. They also ended the year ranked fourth in offensive EPA per play.
Other offenses can make a usually efficient quarterback look inefficient. Aaron Rodgers led the NFL in adjusted EPA per play in 2021 (0.265) but cratered to 21st last season (0.059), better than both Green Bay’s overall EPA per play (0.017) and EPA per dropback (0.035).
This gets into a key distinction within EPA: dropbacks and rushing attempts. Instead of dividing plays between passing attempts and rushing plays, EPA is divided between plays where the quarterback drops back to pass (which can include scrambles and sacks) and those where they hand the ball off. This divide punishes offenses that let their quarterbacks take sacks and rewards those with good pass protection.
One crucial insight of EPA is that running the ball, especially on first and 10, is rarely a recipe for success. Last year, the six teams that ran the ball most often all finished worse than the league average in EPA per play. Only two of them, the Philadelphia Eagles and Baltimore Ravens, made the playoffs. They had to rank first and second in EPA per rush attempt, respectively, to make it there.
The Ravens and Eagles were the only two top-six teams in EPA per rush to get to the playoffs. Five of the top-six teams in EPA per dropback made the playoffs; the lone exception, the Detroit Lions, almost did. Of the top-12 teams in EPA per dropback, 11 made the playoffs. Generally, you’ll want to back teams that pass often and efficiently while ignoring those that commit too many resources to the run.
Using EPA to Evaluate Quarterbacks
We can also use EPA per play to evaluate quarterbacks. This proves especially important when a starting quarterback goes down. No offense can fully compensate for a worse signal-caller, and sometimes a quarterback can look worse in one offense than he did in another, especially if the offensive line can’t protect him.
Using EPA per play as a year-over-year metric to evaluate quarterbacks is a solid approach. A backup quarterback pressed into service in a new offense a year or so after he worked as a starter will likely perform similarly for his new team. Marcus Mariota owned a career EPA per play of 0.09 heading into 2021. He averaged 0.072 in 2022.
Ultimately, the sharpest way to handicap an NFL offense using EPA per play involves a combination of quarterback and team performance. Personally, I use a five-year rolling average that’s weighted for their recent performance to project a given quarterback’s performance. I then combine that with what the offense has been doing this year. That two-pronged approach can help us handicap offenses who are starting their backups.
