NFL Betting Metrics Explained: EPA, DVOA, CPOE, and Success Rate

Consider a real scenario from the 2024 NFL season. A team trails 35-10 entering the fourth quarter. The opponent goes into a prevent defense, sitting in soft zone coverage and conceding underneath routes. The losing team racks up 180 yards and two touchdowns in the final 15 minutes, finishing with 380 total yards and 24 points. The box score looks competitive. The game never was.

Traditional stats treat every yard equally — a 12-yard completion against a two-deep prevent defense in a 25-point blowout counts the same as a 12-yard completion on a critical third-and-10 in a tied game. This is absurd, but it is exactly how total yards, passing yards, and even passer rating work. They are context-free, and context is everything in football.

Game script — the running score differential — shapes every offensive decision. A team that falls behind early abandons the run and throws 50 times. Their quarterback finishes with 340 passing yards and looks great statistically. But the high volume was forced by the deficit, not by offensive quality. Meanwhile, the winning team ran the ball 35 times in the second half with a lead, finishing with only 190 passing yards. Traditional stats say the losing quarterback outplayed the winner. The scoreboard says otherwise.

This distortion directly impacts betting markets. Public bettors overvalue teams with gaudy yardage totals and penalize run-heavy teams that win games 20-13 with “boring” box scores. The advanced metrics we cover in this guide cut through these illusions by measuring efficiency and context rather than volume.

Every metric in this guide shares one principle: not all yards are created equal. A 4-yard gain on 3rd-and-3 is infinitely more valuable than a 4-yard gain on 3rd-and-12. A completion against a top-5 defense is more impressive than the same completion against the worst defense in the league. Modern NFL analytics quantify these differences, producing numbers that actually predict future performance — which is the only thing a bettor should care about.

Every play in football starts from a specific game state: down, distance, field position, and score. Decades of play-by-play data tell us the average number of points scored from each state. For example:

EPA is the difference between the expected points after a play and the expected points before it. If a team starts a play at 1st-and-10 at their own 25 (expected points: +0.9) and completes a 30-yard pass to midfield (expected points: +2.0), the EPA of that play is +1.1. The play added 1.1 expected points.

Imagine two 5-yard runs in the same game:

Traditional stats count both as 5-yard runs. EPA tells you the first run was the equivalent of adding half a point to the team's expected score, while the second run subtracted nearly a full point. This is why EPA is superior to raw yardage for evaluating both plays and players.

When evaluating teams and quarterbacks, the most useful EPA variants are:

Passer rating is a formula invented in 1973 that weights completion percentage, yards per attempt, touchdowns, and interceptions. It assigns no value to context — a 7-yard completion on 3rd-and-6 (first down, drive continues) gets the same credit as a 7-yard completion on 3rd-and-20 (drive over, punt). Passer rating also ignores sacks entirely, rewarding quarterbacks who hold the ball too long and take sacks rather than throwing the ball away.

ESPN's QBR (Total Quarterback Rating) was designed to fix these issues and does incorporate expected points concepts. But QBR is a proprietary black box — ESPN does not publish the full formula, which makes it impossible to audit or replicate. EPA is fully transparent, reproducible from play-by-play data, and available for free. For betting purposes, EPA per dropback is the gold standard for quarterback evaluation because it is open, context-aware, and the most predictive single metric of future offensive output.

DVOA compares every single play to a league-average baseline for the same situation (down, distance, field position, game state). If a team gains 6 yards on 3rd-and-5, DVOA checks: how does that compare to what an average team gains on 3rd-and-5? If 6 yards is above average, the play gets a positive DVOA value.

The “Defense-adjusted” part is crucial. After calculating raw efficiency, DVOA adjusts every play based on the strength of the opponent. If the San Francisco 49ers produce excellent efficiency numbers but played the league's easiest schedule of opposing defenses, their DVOA gets downgraded. If the Baltimore Ravens produce the same raw numbers against the league's hardest schedule, their DVOA gets upgraded. This opponent adjustment is what separates DVOA from simple EPA rankings and makes it far more predictive.

Total DVOA = Offensive DVOA - Defensive DVOA + Special Teams DVOA. A league-average team sits at 0.0%. In a typical season, the best team in the NFL has a total DVOA around +30% to +35%, while the worst team sits around -30% to -35%. The 2023 San Francisco 49ers posted a total DVOA near +33%, reflecting their elite offense and above-average defense. The 2023 Carolina Panthers posted roughly -33%, the worst in the league.

Defensive DVOA is inverted — this trips people up. A defensive DVOA of -20% means the defense is 20% better than average (opponents produce 20% less value than they would against an average defense). A defensive DVOA of +15% means the defense is 15% worse than average. Think of it as: negative is good for defenses because you are subtracting opponent value.

Football Outsiders has published research showing that DVOA is more predictive of future game outcomes than win-loss record, point differential, simple EPA rankings, and Elo ratings. The reason is the opponent adjustment: wins and losses are heavily influenced by schedule difficulty, but DVOA normalizes for it.

A practical example: in 2024, a team might start 7-2 but with a bottom-10 DVOA, having beaten mostly bad teams by narrow margins. Traditional metrics call them a contender. DVOA calls them overvalued — and bettors who faded that team in the second half of the season, when the schedule toughened, found consistent value. Conversely, a 4-5 team with a top-10 DVOA likely lost several close games against strong opponents and is undervalued by the market. This is the core betting edge DVOA provides: identifying teams whose records lie.

Football Outsiders also publishes DAVE (DVOA Adjusted for Variation Early), which early in the season blends current-year DVOA with preseason projections. As the season progresses, DAVE converges toward pure DVOA. For betting purposes, DAVE is more useful in Weeks 1-6 when sample sizes are small, while pure DVOA becomes reliable by Week 8 and is the better metric from midseason onward.

The NFL's Next Gen Stats system tracks every pass using player-tracking data. For each throw, the system estimates the expected completion probability based on:

If the model says a throw had a 45% chance of being completed and the quarterback completed it, the CPOE on that throw is +55 percentage points. If the throw had a 90% chance and the quarterback missed it, the CPOE is -90 percentage points. Averaged across all throws in a season, CPOE tells you whether a quarterback is outperforming or underperforming the difficulty of his attempts.

This distinction is critical for betting. Quarterbacks with high raw completion percentage but negative CPOE are benefiting from scheme — lots of short, easy throws that any NFL quarterback could complete at a similar rate. When those quarterbacks face good defenses that take away the short game, they regress. Quarterbacks with high CPOE maintain their production against tougher coverage because their accuracy is real, not scheme-dependent.

One of the most profitable applications of CPOE is identifying regression candidates:

Research from multiple independent analysts has confirmed that CPOE is more stable year-over-year than raw completion percentage, making it a better predictor of future accuracy. For season-long props and win totals, CPOE should carry more weight than any traditional passing stat.

Success rate simply counts the percentage of plays that meet these thresholds. A team with a 48% success rate is producing a “successful” play on nearly half their snaps — sustaining drives and moving the chains consistently. League average hovers around 44-46% in a typical season. The best offenses push above 50%; the worst fall below 40%.

EPA is influenced heavily by explosive plays — a single 75-yard touchdown pass can swing a team's EPA per play for an entire week. This makes EPA more volatile and harder to project forward. A team that posts +0.20 EPA per play in Week 5 might post -0.05 in Week 6 simply because they hit three explosive plays in the first game and none in the second. Their true offensive quality didn't change; the big-play variance just swung.

Success rate strips out this variance. It doesn't care whether a successful play gained 5 yards or 50 yards — both count as one success. This makes it a measure of consistency and baseline efficiency rather than explosiveness. A team with a high success rate is one that rarely puts itself in 3rd-and-long situations, controls time of possession, and avoids the negative plays (sacks, tackles for loss, penalties) that kill drives.

For betting, this stability translates directly into predictive power. Research from Football Outsiders, The Athletic, and independent analysts consistently finds that success rate from Weeks 1-8 predicts spread-covering performance in Weeks 9-18 better than EPA, DVOA, or win-loss record. The reason is intuitive: consistent teams produce consistent outcomes, and consistent outcomes are what cover spreads.

Like DVOA, success rate is most powerful when you look at both sides of the ball. A team with a 49% offensive success rate and a 41% defensive success rate (opponents are successful on only 41% of plays) is dominating both phases. These teams are the most reliable spread-covering bets in the NFL, because their consistency on both sides makes blowouts more common and close losses rare.

Conversely, a team with mediocre success rates on both sides (45% offense, 45% defense) that has a winning record is likely benefiting from turnover luck, special teams variance, or close-game coin flips. These are classic fade candidates as the season progresses and luck regresses to the mean.

The most direct application is comparing EPA per play rankings to the point spread. If your EPA-based power ratings say Team A should be a 5-point favorite but the sportsbook has them at -3, you have found a 2-point edge. The key is identifying teams whose EPA rankings diverge from market perception.

This happens most often in three situations: (1) teams that had a bad record early but strong EPA, and are now improving their record as luck normalizes; (2) teams that lost a high-profile game that moved public perception more than the underlying metrics warranted; and (3) teams returning from a bye week, where the market sometimes undervalues the rest advantage for teams with already-efficient offenses.

This is one of the most reliable mid-season edges in NFL betting. Every year, several teams have a significant gap between their DVOA ranking and their win-loss record. The market prices games largely based on record — a 7-3 team gets more respect than a 5-5 team regardless of efficiency. DVOA exposes when this respect is misplaced.

Historical analysis shows that teams with DVOA rankings significantly better than their win-loss record tend to cover spreads at a rate above 55% in subsequent weeks, while teams with DVOA significantly worse than their record cover below 45%. This is a persistent, exploitable market inefficiency.

CPOE is the best metric for evaluating quarterback passing props — specifically completion percentage, passing yards, and passing attempts. The logic is straightforward:

Teams with high offensive success rates sustain drives. Sustained drives mean more plays, more time of possession, and more points. When two teams with top-10 success rates meet, the game tends to feature long, methodical drives that push the game toward the over — not because of explosive plays, but because both teams efficiently convert first downs and stay on the field.

Conversely, when two teams with low success rates meet, the game features frequent three-and-outs, quick punts, and fewer total plays. Even if the defenses aren't elite, the offenses' inability to sustain drives limits scoring opportunities. These games often go under.

The nuance: success rate predicts play volume, which predicts totals. A game between two high-success-rate teams will have more total plays (and thus more scoring opportunities) than a game between two low-success-rate teams. Oddsmakers account for offensive and defensive talent but sometimes underweight the play-volume effect of success rate differentials.

PROE measures how often a team passes the ball compared to what is expected given the game situation — specifically in neutral game scripts (when the score is close enough that game script isn't forcing the play call). Teams that pass more than expected in neutral situations tend to score more points, because the modern NFL passing game is inherently more efficient than the running game.

For totals betting, PROE is a leading indicator. A team with a high PROE is choosing to pass even when they don't have to — this typically means their coaching staff believes their passing attack is their primary weapon, and they are willing to deploy it aggressively. When two high-PROE teams meet, the increased passing rate leads to more explosive plays, more first downs, and more scoring. This is a strong overs signal, particularly when the total is set lower than expected based on the teams' overall efficiency rankings.