DVOA Explained: Football Outsiders’ Long Shadow on Modern NFL Analytics

January 14, 2007. Indianapolis Colts at Baltimore Ravens, divisional round of the playoffs. The Colts win 15-6 on five Adam Vinatieri field goals, an offense that famously could not score a touchdown against the league’s best defense, and Peyton Manning is later credited as the architect of a win that, on the scoreboard, looks like a wash. At the same hour, in a Brooklyn apartment, a software engineer named Aaron Schatz is updating the season-end DVOA leaderboard at Football Outsiders. His model, trained on the previous season’s play-by-play, has Indianapolis a clear top-three team despite that day’s offensive performance. Baltimore’s defense, the model says, was always going to suppress them. The number had told the story before the broadcast did, and the writers who read it that morning were already two analytical steps ahead of the talk-radio cycle that would, by Wednesday, have rebuilt Peyton Manning’s reputation around a single playoff win.

DVOA — Defense-adjusted Value Over Average — was, for fifteen years, the most influential single metric in public NFL analysis. It predated Brian Burke’s EPA work by half a decade, predated nflfastR by a decade more, and shaped how a generation of writers thought about football efficiency before the play-by-play data was free, the models were open-source, or the conversation had spilled out onto Twitter. The metric is no longer the public-facing default; Football Outsiders the company has paused, Schatz has moved on, and EPA has become the lingua franca of the modern football conversation. But DVOA’s long shadow is unavoidable. It taught the public that football efficiency could be measured. It taught writers that opponent adjustment mattered. It taught the league that play-by-play, not box scores, was the actual ledger. The metric’s specific math now lives in retirement; its ideas are, in 2026, embedded in everything serious anyone writes about the NFL.

I have been writing about football analytics since 2014, mostly through the lens of DVOA’s successors, and the metric I keep finding myself coming back to as the reference point for how the public-facing analytics conversation got built is the one this article is going to unpack. DVOA, where it came from, how it actually worked, where it broke, and why every modern NFL analytical writer still owes the framework a measurable debt, is the subject of this article.

The origin: where DVOA came from

DVOA was created by Aaron Schatz in the early 2000s, initially as a side project on his football statistics blog that eventually became Football Outsiders. Schatz’s training was in journalism, not statistics, but the methodological insight he stumbled into in 2003 — that every football play could be evaluated against a league-average baseline for the same situation — became the foundational frame for two decades of NFL analysis.

The technical inspiration came from baseball. The Bill James sabermetric tradition, by the early 2000s, had produced metrics like OPS+ and ERA+ that normalized player performance against league average, adjusted for park and era. Schatz’s question was whether the same logic could apply to a sport where the play-by-play data was, at the time, considerably messier and the relationships between individual plays and team outcomes were less easily isolated. The early DVOA work spent considerable energy on the data plumbing — getting clean play-by-play from STATS Inc., normalizing across teams, building reasonable expected-value baselines for every down-and-distance situation in the league.

The first DVOA leaderboards published in 2003 and 2004 produced rankings that, in retrospect, looked nothing like the mainstream consensus of the time. The model was high on a then-emerging Patriots dynasty, low on traditional offensive powerhouses whose yardage stats inflated their reputations, and surprisingly accurate at predicting playoff outcomes. The Football Outsiders Almanac, published annually starting in 2005, became the closest thing the NFL had to a Baseball Prospectus equivalent — a season-preview reference that the analytics-curious portion of the league office, the gambling community, and the small but growing number of stat-aware writers all read seriously.

By the late 2000s, DVOA was being cited in mainstream coverage with increasing regularity. ESPN began incorporating it into game previews. Sports Illustrated profiled Schatz. The metric’s influence on NFL coverage was, by 2010, large enough that the methodology — opponent-adjusted per-play efficiency, expressed as a percentage above or below league baseline — had become the default conceptual frame for any serious analytical conversation about football. EPA, when it emerged later as a competing public-facing metric, was working in a conceptual space DVOA had already cleared.

How DVOA works: in plain language

DVOA’s mechanics are straightforward in concept and complex in execution. Every play in the NFL season is assigned a value based on its situational context (down, distance, yard line, time, score) and its outcome (yardage gained, first down converted, turnover, penalty). That value is compared to the league-average outcome for the same situation. The difference — the value above or below the baseline — is expressed as a percentage. A play that gained 1.20× the league average for that situation is +20% DVOA. A play that gained 0.80× is -20%.

Sum the per-play values across a team’s offense, defense, or special teams over a season, weight them by play frequency, adjust for opponent quality, and the result is a single DVOA percentage for each unit. An offense at +20% DVOA is twenty percent more efficient than the league average against the same schedule of defenses. A defense at -20% DVOA suppresses opponents to twenty percent below the league average. Special teams DVOA is measured similarly but on a smaller play volume.

The opponent adjustment is iterative. The first pass produces a raw DVOA for every team. The model then re-evaluates each team’s performance against the new ratings of their opponents, which propagates a second round of adjustments. The process repeats until the ratings stabilize. The output is a DVOA score that has, in principle, already accounted for the quality of the defenses each offense faced and the quality of the offenses each defense faced.

The single most important thing to understand about DVOA is that it is a percentage, not a points-per-play number. EPA expresses efficiency in expected-points units; DVOA expresses it as a fraction above or below baseline. The two metrics measure overlapping things but in different units, which is part of why the public conversation occasionally talked past itself when comparing them.

The critical component: opponent adjustment, iterated

The single most influential conceptual contribution of DVOA to NFL analysis was the insistence on opponent adjustment. Raw per-play efficiency, without accounting for the quality of opposition, was — by the late 1990s — already a known-but-underutilized public stat. What DVOA did was insist that the adjustment had to be iterative and that the iteration mattered.

The math of iterative adjustment is intuitive once you see it. Imagine ranking teams by their record alone. The Patriots are 14-2; the Colts are 13-3. By raw record, the Patriots are slightly better. But the Patriots’ wins came against a softer schedule; the Colts beat three top-five defenses. A simple opponent adjustment would credit the Colts for the schedule. The iterative version says: wait, we now know the Colts are slightly better than we thought before the adjustment, which means the teams the Colts beat — and the teams that beat the Colts — should also be re-rated upward and downward respectively. After a second pass, the picture sharpens. After a third or fourth, it usually stabilizes.

DVOA vs the alternatives: a comparison

The metric exists in a now-crowded landscape of public NFL analytics. A short comparison:

The honest version of modern NFL writing references multiple frameworks. DVOA’s historical archive remains the longest-running consistent public dataset for football efficiency. EPA, with its open-source pipeline and easy reproducibility, has become the practical workhorse. Success Rate fills the consistency gap. PFF fills the player-evaluation gap. The writer who reads two or three of these together is, more often than not, producing the most defensible analysis.

What the data needs: inputs

DVOA, in its full form, requires play-by-play data with situational context — down, distance, yard line, time remaining, score, play type, outcome — for every snap of every game in the NFL season. The data has to be cleaned for inconsistencies in play categorization (sacks counted against passing offense, end-of-half kneel-downs filtered out, garbage time partially or fully removed depending on the version). Football Outsiders built and maintained the cleaning pipeline internally for two decades, which is part of why the metric’s reproducibility was always limited to the team behind the site.

The model also needs a reasonable baseline for each situation. DVOA’s baseline was constructed from twenty-plus years of historical NFL plays, which means the expected-value calculations reflect the league as it was during the training window. As the NFL has shifted toward more passing, more spread formations, and more fourth-down aggression, the baselines have had to be retrained periodically — a maintenance task that, in retirement, the metric no longer receives.

For writers in 2026 who want to do the DVOA-style analysis without the original tool, the closest equivalent is to compute opponent-adjusted EPA per play from nflfastR data. The mechanics are similar; the output is denominated in expected points rather than percentage above baseline. The conceptual frame is the one Schatz pioneered.

Building the analysis: a working framework

The practical workflow for using DVOA-style thinking in modern NFL writing, even when DVOA itself is no longer being updated:

1. Pull EPA per play data for the relevant team and season, via nflfastR or RBSDM.com.
2. Apply an opponent adjustment. If the team in question has played a soft schedule, their raw EPA per play overstates their quality. Pull the average EPA per play allowed by their opponents’ defenses, league-wide. Adjust the team’s number downward by the differential.
3. Compare to the league baseline. The league average EPA per play hovers near zero. A team that, after adjustment, is at +0.10 EPA per play is roughly equivalent to a +15% offensive DVOA team in the older framework.
4. Check consistency with Success Rate. A team with high EPA but low Success Rate is feast-or-famine. A team with both is a problem for opposing coordinators.
5. Cross-reference with film and beat reporting. The numbers tell you which teams to take seriously. The film tells you why they’re succeeding.

Where this gets weird: common mistakes

DVOA, and DVOA-style frameworks, have known failure modes.

Garbage-time pollution was a perennial issue. DVOA’s various filters for garbage time changed over the years, and writers who quoted the metric without specifying which filter were sometimes describing different underlying numbers. EPA-based analysis inherits the same problem.

Single-game DVOA was noisy. A team posting +50% DVOA in one game and -30% in the next was, in most cases, telling a story about the games rather than about the team’s true talent. The Football Outsiders Almanac was always clearer than the weekly numbers on this point; the public conversation was less so.

Quarterback DVOA inherited the offensive line problem. A QB’s per-play efficiency is a joint product of the QB and the offensive line and the receivers and the play-caller. DVOA’s per-player splits attempted to isolate individual contribution but, like every public metric in the football-analytics era, struggled with the noise. Modern PFF and EPA-CPOE blends do better, marginally.

The opaque methodology was a real cost. Football Outsiders never fully open-sourced the DVOA model. The exact weighting of situational variables, the precise opponent-adjustment algorithm, and the handling of edge cases were inside the firm. That made replication difficult and limited the academic community’s ability to extend the work.

When DVOA-style thinking shines: use cases

The applications where the framework has earned its long career:

Mid-season team evaluation. A team’s record at the midway point is a function of underlying efficiency plus turnover luck plus opponent strength. DVOA-style thinking — opponent-adjusted per-play efficiency — produces a much cleaner read on actual team quality than the standings alone. The 2007 Patriots, 2017 Eagles, and 2022 Lions were all teams the DVOA-style frame identified as elite well before the mainstream narrative caught up.

Bowl-game-style projections. When a playoff matchup pits an offense from one conference against a defense from another, raw per-play numbers can be misleading because of opponent-quality differences. Opponent-adjusted frameworks normalize the comparison. DVOA had a strong historical record in playoff prediction during its peak years.

Identifying overperforming and underperforming teams. A team whose record substantially exceeds their DVOA-derived efficiency is, on average, going to regress. A team whose record undershoots their efficiency is, on average, going to improve. The same regression logic applies to modern EPA-derived numbers.

Long-form profiling. The Football Outsiders Almanac, at its best, used DVOA as the analytical spine of a long-form preseason preview that combined statistics with reporting. The format was inherited by The Athletic, Football Outsiders’ own later years, and serious football podcasts. The format works because the underlying metric was disciplined enough to support the prose.

A working example: the 2017 Philadelphia Eagles

The 2017 Philadelphia Eagles are one of the cleaner DVOA case studies of the metric’s peak era. The team finished the regular season 13-3 with the league’s best DVOA at +30.4%, a number that, in real time, was being treated by many in mainstream coverage as inflated by a soft NFC East schedule. The Football Outsiders Almanac and Schatz himself argued that the opponent adjustment had already accounted for that, and that the Eagles were genuinely the most efficient team in the league. The model further predicted that Carson Wentz’s late-season knee injury, replacing him with Nick Foles, would degrade the offense by roughly six to eight percent DVOA points — meaningful but not catastrophic. The Eagles, with Foles, beat the Falcons, the Vikings, and the Patriots in succession to win Super Bowl LII. The Vinatieri-of-2007 redux moment for the model was that it had been close to correct about Foles’s likely impact, and the Eagles’ underlying offensive identity remained near-elite even with the backup quarterback.

The Eagles’ run produced one of the better-aged retrospective pieces of football analytics writing, because the model had said something specific and falsifiable before the playoffs and the playoffs confirmed it. That kind of analytical prediction, properly bounded, is what DVOA — and the EPA-based work that followed — can do at its best. It is also what most mainstream coverage in 2026 still fails to do consistently.

The limits: what DVOA could not tell you

The honest version of this writing names the limits.

DVOA could not tell you who was going to win on Sunday. It told you who had been the better team in the games already played, and it produced a calibrated probability for upcoming games based on those readings. The translation between season-long efficiency and single-game outcomes is non-trivial in any sport, and the NFL’s high variance per snap and per drive meant that even a strong DVOA favorite could lose to a strong DVOA underdog with regularity.

DVOA could not capture individual contributions cleanly. The team-level numbers were strong; the per-player splits were noisier than the public coverage often acknowledged. Quarterback DVOA, in particular, mixed the QB’s actual contribution with the offensive line’s and the receivers’ in ways the metric could not fully unpack.

DVOA could not, in retirement, keep up with the modern league. The metric’s last fully-updated season was 2023. The NFL’s continued evolution — the rules changes around quarterback hits, the shift in three-receiver personnel rates, the rise of pre-snap motion — required ongoing recalibration that, post-Football Outsiders, no one was doing publicly under the DVOA brand.

One additional limit worth naming. DVOA’s transparency was always limited by the proprietary nature of Football Outsiders’ implementation. Even at its peak, writers who wanted to extend the framework or apply it to college football, the CFL, or other contexts had to build their own version from scratch. The intellectual influence of the metric was enormous; the direct reusability was limited. The next generation of opponent-adjusted models — at PFF, at ESPN, in the cfbfastR ecosystem — built on the conceptual foundation rather than the actual code, which is part of why the public NFL analytics conversation in 2026 looks structurally similar to 2010 even as the specific tools have rotated.

Frequently asked questions

What happened to DVOA and Football Outsiders?

Football Outsiders the company paused its primary operations in 2024 after a series of ownership changes. The DVOA leaderboards stopped being updated in real time. Aaron Schatz has moved on to other projects, including ESPN columns and his own newsletter. The historical DVOA archive remains accessible and is still cited regularly in NFL writing, but the metric is no longer in active development.

Is DVOA still useful even though it’s not being updated?

The historical archive is still valuable for cross-era comparison and for retrospective analysis. The conceptual frame — opponent-adjusted per-play efficiency — is embedded in every serious public NFL metric that exists now. Modern writers who want a DVOA-style number for current seasons typically build it themselves from nflfastR data, using EPA per play as the input rather than the proprietary value-over-average that Schatz built.

Why was DVOA expressed as a percentage and not in points?

Schatz’s original framing came from the baseball sabermetric tradition (OPS+, ERA+), where league-relative percentages were the convention. The percentage made cross-era comparison clean — a team at +30% DVOA in 2007 is comparable to a team at +30% DVOA in 2019, regardless of league scoring environment. EPA, by contrast, is denominated in expected points, which makes it more intuitive for individual play evaluation but harder to compare across years when scoring environments shift.

What’s the best DVOA-equivalent for 2026 NFL analysis?

For most public-facing analytical work, opponent-adjusted EPA per play (computed from nflfastR data) is the closest functional substitute. The methodology requires more technical work than DVOA did — DVOA presented the adjustment baked into the headline number — but the conceptual frame is the same. PFF’s grade-based system serves a different but complementary purpose for player evaluation.

Sources and further reading

• Football Outsiders archive — the long-running home of DVOA, still useful for historical reference and the Almanac series.
• Aaron Schatz’s current writing — the founder of DVOA continues publishing football analysis in newsletter form, with the same methodological discipline.
• nflfastR documentation — the open-source R package that enables modern DVOA-style analysis using EPA as the input.
• Pro Football Focus — the leading commercial alternative to DVOA-style team rating, with deeper player-level grading.
• RBSDM.com — clean public-facing EPA leaderboards with opponent-adjustment options.

The Schatz-and-Indianapolis story that opened this article — 2007 divisional round, the Vinatieri five-field-goal game, the model that had read the matchup correctly before kickoff — is, in retrospect, the kind of moment the public analytics conversation in football was always reaching toward. The metric is in retirement. The intellectual debt is permanent. Every careful piece of NFL writing in 2026 inherits, knowingly or not, the methodological frame that one journalist-turned-analyst built in his Brooklyn apartment in 2003. For the conceptual companion to this piece — how the modern public-facing equivalent works — our guide to EPA is the natural next read.