How ESPN's Football Power Index (FPI) Calculates NFL Win Projections

As the 2026 NFL season kicks off, the broadcast graphics and digital previews are once again dominated by a single, decimal-point metric. On Tuesday, ESPN's Football Power Index (FPI) projected a struggling campaign for the Arizona Cardinals, while simultaneously driving betting market splits across college football. The metric operates as the definitive mathematical baseline for the network's coverage, generating the win probabilities that hover on the screen before every kickoff. Yet the mechanics beneath that percentage remain largely opaque to the average viewer.[1]

FPI is fundamentally a measure of team strength, expressed as the number of points a franchise would beat an average opponent by on a neutral field. If a team carries a +2.6 rating, the model expects them to win by roughly a field goal against a league-average roster in a vacuum. To generate that single number, the system isolates a team's offensive, defensive, and special teams units, evaluating each through the lens of Expected Points Added (EPA).[2]

EPA serves as the foundation of the entire architecture. Every single play in a football game is graded against historical data to determine how much it increased or decreased a team's likelihood of scoring. A three-yard run on first-and-ten might carry a negative EPA, while a three-yard run on third-and-two carries a highly positive one. By aggregating these play-by-play values, the model strips away the luck of fumble recoveries or dropped passes to measure true down-to-down efficiency.[2][3]

The NFL version of the index diverges significantly from its college football counterpart by explicitly isolating one individual position: the quarterback. Because the professional game is so heavily dictated by the player under center, the quarterback is the only roster spot that directly alters the math. The model adjusts the baseline rating based on whether the starter is active, factoring in injuries, rookie variance, and the historical drop-off to the specific backup on the depth chart.[2]

Environmental and logistical variables are then layered over the efficiency ratings to create a game-specific projection. The NFL FPI accounts for the altitude of the stadium, which disproportionately affects games played in Denver, as well as seasonal weather effects as the schedule stretches into December and January.[2]

Travel and rest also mathematically shift the odds. According to the model's historical weighting, extreme cross-country travel—such as the Seattle Seahawks flying to face the Miami Dolphins—is worth approximately half a point to the home team, assuming all other factors are equal. Similarly, a team operating on extra days of rest receives a quantifiable bump, particularly when facing an opponent coming off a short week.[2]

Early in the season, the model relies heavily on a Bayesian framework anchored by preseason expectations. These initial priors are heavily influenced by Las Vegas win totals and betting market consensus. "FPI's preseason predictive ratings are primarily based on win totals from the betting market in conjunction with each team's schedule," ESPN's methodology documentation explains, noting that these priors prevent early-season anomalies from wildly skewing the ratings.[3]

As the season progresses, a "K factor" dictates how quickly the ratings react to new results. The model slowly phases out the preseason priors, replacing them with actual on-field EPA data. The margin of victory plays a critical role in this updating process, though the algorithm applies a declining marginal benefit to blowouts to prevent a single 40-point victory from breaking the scale.[3]

Once the baseline ratings and game-specific adjustments are locked in, the system turns to forecasting the future. The model simulates the remainder of the NFL season 10,000 times using a Monte Carlo method. "We use these ratings to simulate the season thousands of times, with the results forming our projections," the network's developers note. Those 10,000 simulated universes produce the percentage chances of a team winning its division, securing a playoff berth, or reaching the Super Bowl.[2]

In June 2026, for example, the model ran its preseason simulations and handed the Los Angeles Chargers a 65.1 percent chance to make the playoffs, ranking them fifth in the AFC. The system calculated a 4.4 percent probability that Jim Harbaugh's squad would win the Super Bowl, translating the team's +2.6 baseline rating into a concrete postseason forecast. "We're hungry. I think everyone who's been here for a while knows we're due," Chargers tackle Rashawn Slater remarked regarding the team's expectations.

Despite its mathematical rigor, the index frequently draws intense scrutiny when its underlying efficiency metrics clash with the reality of the standings. In November 2025, the model faced widespread criticism after ranking the 2-8 New York Giants 15th overall—one spot behind the 8-2 New England Patriots, and five spots ahead of the 6-3 Chicago Bears team that had just defeated them.

That disconnect highlights the philosophical divide at the heart of sports analytics. The model does not care that the Giants lost; it cares that their underlying down-to-down efficiency suggested they were playing better football than their record indicated. For algorithmic forecasters, that predictive accuracy is the entire point. In 2016, FPI favorites won 73 percent of their regular-season games, outpacing the closing lines in Las Vegas.[2]

For traditionalists, however, a model that ranks a two-win team ahead of an eight-win team fails the basic logic test of competitive sports. The debate centers on whether a team's value is defined by its theoretical capability or its actual results. As the 2026 season unfolds, the FPI will continue to update every Tuesday morning, adjusting its 10,000 simulations based on the latest data, regardless of whether the public agrees with the math.[1][4]