NFL player stats, 2009 to 2025: passing peaked, defenses changed

About the data

Where the numbers come from and how they are computed

Source. The nflverse-data stats_player release: weekly player stats, one file per season, licensed CC-BY-4.0. Credit to the nflverse project and its contributors. Definitions of the advanced columns come from the nflverse player stats data dictionary.

What one row is. One row is one player in one game. The cleaned data has 304,515 rows and 69 columns for 17 seasons (2009 to 2025), 8,093 players (8,081 in the regular season) and 32 teams. The time column is season (and week) and the group columns are player and team. This report and the dashboard's starting view use regular-season rows only, which is 291,750 of them; playoffs can be added with the dashboard's season-type filter.

Rows dropped and other cleaning. The source files hold 304,877 rows. 362 were dropped for having no player (below), 0 for having no opponent and 0 as duplicates, which leaves 304,515 rows.

• 362 rows dropped for having no player. These rows have no player ID, no player name and no position, so they cannot be a player-game. There is exactly one such row for each of the 362 season-weeks in the source (21 to 22 per season, 294 in the regular season and 68 in the playoffs), not one per team per week. Each has a team code, and 337 of them have non-zero stats, almost all penalties (330 rows) and safeties (84 rows). The source does not say what they are; they look like penalties and other plays charged to a team instead of a player. What was lost, counting regular-season totals in the dropped rows out of all rows: sacks 11 of 20,708, interceptions 0 of 7,521, solo tackles 163 of 353,204, QB hits 6 of 47,631, passing yards 0 of 2,163,026, rushing yards 0 of 1,016,067, receptions 2 of 191,313. Penalties are the one number that is visibly affected: 2,764 of 55,658 (5.0%) sit in dropped rows, so penalty rates on the dashboard and the team field are about 5% lower than the source's total. Nothing in the findings above uses penalties.
• 4,536 all-zero rows kept. 4,536 rows (4,342 of them in the regular season) have a player ID, name, position and team but no tracked stat in the columns kept here (mostly linemen, defensive backs and linebackers who played without recording a stat). They are real players who appeared in the game (1.5% of regular-season rows), so dropping them would understate who played. They add nothing to any sum, and they do count as player-games in the "Player-games" measure and in per-player-game averages.
• Team codes. The source uses one code per franchise across moves: the Raiders are LV, the Chargers LAC and the Rams LA in every season, so there are 32 teams in all 17 seasons and no renaming was needed.

Tackles for loss are not recorded in 2009 to 2011 and are marked NA (not zero) there. More generally, NA means the source has no value for that player in that game (for example, no pass attempts means no passing EPA); sums skip NA. Player names are not unique, so players are identified by player_id. Full details are in data/DATA_QUALITY.md and data/DATA_DICTIONARY.md.

Terms in plain words.

EPA (expected points added)

How many points a play was worth compared with the average play in the same situation. Example with made-up round numbers: a drive worth 1.0 expected points before a pass and 2.0 after a 20-yard catch gives that pass an EPA of +1.0. The values come from nflverse's model, not from this project.

Team-game

One team in one game. A season of 17 games is 17 team-games for that team.

Sack

A tackle of the quarterback behind the line of scrimmage before he throws. Defenders can share one (each gets half).

QB hit

A defender hits the quarterback as or just after he throws, without a sack.

Carry

One running play: a player runs with the ball (a rush attempt).

Dropback

A pass attempt or a sack: every time the quarterback drops back to pass. EPA per dropback is passing EPA divided by attempts plus sacks.

CPOE (completion percentage over expected)

How much more or less often a passer completes passes than nflverse's model expects given how hard each throw was, in percentage points. +2 means two points better than expected.

Opportunities

The plays a player was directly involved in: pass attempts + sacks + carries for a quarterback, carries + targets for a running back, receiver or tight end. For quarterbacks, "EPA per play" in finding 12 means EPA per opportunity.

Replacement level

What the players just below the starters produced: this site's rule puts everyone ranked below the top 32 quarterbacks (64 receivers) by opportunities in the replacement pool. The pool includes benched and injured players, so it is a low bar.

EPA over replacement (EPAOR)

A player's total EPA minus what the replacement rate would have produced on the same number of opportunities.

Bell-cow

A running back who gets most of his team's carries.

How every number is computed.

• Team-game: one team in one game. Per-team-game rates add up every player's stat for the season and divide by the number of regular-season team-games (32 teams, 16 games each through 2020, 17 from 2021, with one fewer game in 2022).
• Player-season counts (4,000-yard passers, 1,000-yard rushers, 300-carry backs, 100-catch receivers, players with 6+ interceptions): each player's regular-season total for one season is compared to the threshold, and players who met it are counted once. A player who changed teams has one total across teams.
• Interceptions per team-game is defensive interceptions divided by team-games.
• Sacks per pass attempt is defensive sacks (halves count) divided by pass attempts. Sacks are not part of attempts.
• QB hits and sack leaders are single-season player totals. Ranking ties are broken by player name. No two players tied for the sack lead in any season.
• Sack rates (two different ones): finding 8's "sacks per pass attempt" is defenders' sacks (halves count) divided by pass attempts, for the whole league. The dashboard's "Sack rate (sacks taken per dropback)" is sacks suffered by quarterbacks divided by attempts plus sacks, so the two are not the same number and are not expected to match.
• Percent change: 1 minus the later value divided by the earlier value (for example 2025 against the peak, or against 2009).
• Top 32 and top 64 "starter" rate: the starters' total EPA divided by the starters' total opportunities in that season, one pooled rate, like the replacement rate. The "averaged" figures in finding 12 are simple means of the 17 season rates, not one rate pooled over every play.
• Players who changed teams: a player's season total is one number across teams. In the charts his block is split by team in proportion to each team's share of the stat (for EPAOR, each team's share of his opportunities).
• Advanced stats on the dashboard: EPA per dropback is passing EPA divided by attempts plus sacks. ANY/A is (passing yards + 20 x touchdowns - 45 x interceptions + sack yards lost) divided by (attempts + sacks), with sack yards stored as negatives. CPOE is each game's CPOE weighted by that game's attempts. RACR is receiving yards divided by receiving air yards. Target share, air-yards share and WOPR are averages over games in which the player was targeted. Each dashboard measure states its definition under the chart.
• EPA over replacement (EPAOR, this site's version): regular season only, one position group at a time (quarterbacks, running backs, wide receivers, tight ends). Opportunities are attempts + sacks + carries for quarterbacks and carries + targets for the others. Each season players are ranked by opportunities (ties by player ID); the top 32 quarterbacks, 32 running backs, 64 wide receivers and 32 tight ends are the starters and everyone below them is the replacement pool. The replacement rate r is the pool's total EPA divided by the pool's opportunities, and EPAOR = a player's EPA - r x opportunities. The cutoffs are a judgment call, not nflWAR's. Scores are in expected points, not wins, and are only comparable within a position. A pass's EPA appears under both the passer and the receiver, so EPA must not be added across players.

Limits. Totals are nflverse's numbers and can differ slightly from official league totals. Stats are only as good as the source; tackles depend on stat-crew scoring and are not used here. Metrics that need play-by-play, Next Gen Stats or Pro Football Reference data (EPA allowed, pressure rate, time to throw, missed tackles) are not in this dataset. Reasons for the trends, such as rule changes, are outside the data and are not tested here.

Every number on this page is computed by scripts/build_report.py from data/seasons/, and scripts/check_numbers.js recomputes them with the dashboard's own code.