How Football ChangedThe long arc
NFL Offenses Got Better on Third Down Around 2014. Since 2020, Third Downs Have Been Shorter.
NFL offenses converted 39.4% of third downs in 1999–2001 and 42.3% in 2015–25. At the same distances, the gain came in one step around 2014; after it, the rate at any given distance held near that level, slipping slightly in recent seasons. From 2020, shorter third downs made up the slip and kept the overall rate level, and third-and-long is as hard as it was in 1999. The biggest change came after third down failed: more teams went for it on fourth, and more of them made it.
By Edgehalla ResearchPublished 17 min read
On Dec. 2, 2012, the Arizona Cardinals faced third down 15 times against the New York Jets, with Ryan Lindley at quarterback. They converted none of them and lost 7–6.
No team in play-by-play data back to 1999 has faced more third downs in a game without converting one. Four years earlier, Drew Brees’s New Orleans Saints went 11 for 11 in a 42–7 win over Detroit, the only perfect day in our data on eight or more tries.
Between those two afternoons sits the rest of the league, which converts more third downs than in 1999. Did offenses get better at third down, or just stop facing third-and-long?
On this page
Offenses did get better at third down, in one step around 2014
The first answer is yes, once. In 1999, NFL offenses converted 39% of their third downs. In 2025 they converted 42%. (A conversion here is a first down or a touchdown on the play, or a first down handed over by a defensive penalty.) That looks like a slow climb. But a third-and-2 is easier than a third-and-9, so a season full of short third downs flatters the rate. In 2025, 46% of third downs needed 7 or more yards, against 50% in 1999–2001. So we grade every season on the 1999–2001 mix. We take how often teams converted third-and-2, third-and-9 and every other distance that year, and weight those rates as if they had faced the old mix of distances. Graded that way, a season earns nothing for an easy run of third-and-2s. The same-distance rate sits flat near 40% through 2013, steps up in 2014, and has averaged about 41.5% since. The newest seasons sit a little below that average. In 2022–25 the same-distance rate averaged 40.8%, about level with 2002, the best season before the step.
At the same distances, NFL third-down conversion stepped up once, around 2014
The bold line holds the mix of distances fixed at its 1999–2001 level, so it moves only when teams converted the same distances more or less often. Shaded: where each season’s true figure most likely sits, given how far luck can push one season. Dashed: the level before and after 2014, the break the data picks.
Show data
| Year | At 1999–2001 distances | At 1999–2001 distances (range) | Actual rate | 1999–13 level | 2014–25 level |
|---|---|---|---|---|---|
| 1999 | 39.0% | 37.9% to 40.1% | 38.6% | 40.0% | – |
| 2000 | 40.2% | 39.1% to 41.3% | 40.3% | – | – |
| 2001 | 38.9% | 37.8% to 40.0% | 39.2% | – | – |
| 2002 | 40.8% | 39.7% to 41.8% | 41.3% | – | – |
| 2003 | 39.4% | 38.3% to 40.4% | 39.7% | – | – |
| 2004 | 39.6% | 38.5% to 40.7% | 40.0% | – | – |
| 2005 | 40.7% | 39.6% to 41.8% | 40.6% | – | – |
| 2006 | 39.9% | 38.8% to 41.0% | 40.7% | – | – |
| 2007 | 40.2% | 39.1% to 41.3% | 41.3% | – | – |
| 2008 | 40.2% | 39.1% to 41.3% | 41.8% | – | – |
| 2009 | 39.8% | 38.7% to 40.9% | 40.8% | – | – |
| 2010 | 39.9% | 38.8% to 41.0% | 40.4% | – | – |
| 2011 | 40.4% | 39.3% to 41.5% | 40.6% | – | – |
| 2012 | 40.7% | 39.6% to 41.8% | 41.2% | – | – |
| 2013 | 40.2% | 39.1% to 41.3% | 40.8% | 40.0% | – |
| 2014 | 42.4% | 41.3% to 43.5% | 42.8% | – | 41.5% |
| 2015 | 41.4% | 40.3% to 42.5% | 41.8% | – | – |
| 2016 | 41.8% | 40.7% to 42.9% | 42.3% | – | – |
| 2017 | 41.4% | 40.3% to 42.5% | 41.5% | – | – |
| 2018 | 41.4% | 40.3% to 42.5% | 42.0% | – | – |
| 2019 | 42.2% | 41.0% to 43.3% | 42.2% | – | – |
| 2020 | 42.8% | 41.6% to 43.9% | 44.6% | – | – |
| 2021 | 41.6% | 40.5% to 42.7% | 43.1% | – | – |
| 2022 | 40.5% | 39.4% to 41.5% | 42.1% | – | – |
| 2023 | 40.6% | 39.5% to 41.6% | 41.5% | – | – |
| 2024 | 41.2% | 40.1% to 42.3% | 42.3% | – | – |
| 2025 | 40.9% | 39.9% to 42.0% | 42.2% | – | 41.5% |
- 2020: 2020 (outlier)
We let the data choose the year. We tried every season from 2002 to 2023 as the start of a new level, and 2014 fit best. Then we shuffled the seasons into 2,000 random orders, and none of those made-up histories produced a step as sharp, so the step is too big to be luck. The exact year is less certain. When we redraw the games at random and run the search again, 2014 comes out on top about 62% of the time, and the year most likely lands somewhere from 2010 to 2014. A steady climb fits worse. The seasons before 2014 sit flat, then jump. The data finds no second step since. Seasons after the step have bounced around a little more than luck usually makes them. Luck alone would make them bounce that much about one time in ten.
2014 is also the season officials set out to call more defensive holding and illegal contact. That August, referee Ed Hochuli told the Milwaukee Journal Sentinel (via Pro Football Talk): “The big one is holding.” He went on: “We’ve also tightened up the rule on illegal contact.” One penalty tracker counted defensive-holding flags rising from .678 a game in 2013 to .938 in the first week of 2014, a one-week sample, as the site noted. Our story on penalty flags follows those emphasis years.
Penalty first downs make up part of the step: count only plays and the step shrinks from about 1.5 points to about 1.0. On third down, defensive penalties that handed over a first down did become more common. They came about 3.0 times per 100 third downs in 1999–2001 and the same in 2006–12, then 4.1 in 2013–19 and 3.7 in 2021–25. But they were already as common in 2012 as in 2014: 4.2 per 100 in 2012, 3.5 in 2013 and 4.1 in 2014. So they don’t line up neatly with the step. Count only first downs gained on the play itself, and the same search still picks 2014, with a smaller step, from 38.0% to 38.9%. On that count the year is less certain. Redraw the games and run the search again, and the year most likely lands between 2004 and 2014, on 2014 itself about 46% of the time. The numbers date the step. Flags, rule changes and better play all landed around then, and play-by-play data can’t pull them apart.
One season sits above the rest. In 2020 offenses converted 45% of 6,663 third downs, the highest rate in our data, and 43% at the same distances, also the highest. Then it fell back. The red-zone story found the same hot season and no way to say why. This season looks like the rest of the post-2014 years: in weeks 1–6 of 2026, the complete weeks when our data was pulled on Oct. 26, 2026, offenses converted 42.5% of 2,471 third downs, against 42.6% in the same weeks of 2015–25.
From the 2010s to the 2020s, the only gain came from shorter third downs
That step is one part of the rise. The other part needs no better play at all: if offenses face shorter third downs, the rate climbs with nobody converting any better. So the rise splits in two, a shorter mix of distances and better conversion at the same distance. From 1999–2001 to 2015–25 the rate rose about 3 points, meaning about 3 more conversions for every 100 third downs. Shorter distances supplied about 1 of those points (29% of the rise). Better conversion at the same distance supplied about 2. (Each part is rounded on its own; the table below has the exact split.)
From 1999 to 2025, shorter third downs alone would have added about 1.6 points, mostly after 2019
Actual third-down rate, against the rate each season would have had with its own mix of distances converted at 1999–2001 rates. This compares single seasons; the text compares groups of seasons (1999–2001 against 2015–25), where shorter distances add about 0.9 points, because 2015–25 still includes the longer third downs of 2015–19.
Show data
| Year | Actual | Distance mix only |
|---|---|---|
| 1999 | 38.6% | 39.0% |
| 2000 | 40.3% | 39.5% |
| 2001 | 39.2% | 39.6% |
| 2002 | 41.3% | 39.9% |
| 2003 | 39.7% | 39.8% |
| 2004 | 40.0% | 39.8% |
| 2005 | 40.6% | 39.4% |
| 2006 | 40.7% | 40.2% |
| 2007 | 41.3% | 40.4% |
| 2008 | 41.8% | 40.7% |
| 2009 | 40.8% | 40.3% |
| 2010 | 40.4% | 39.8% |
| 2011 | 40.6% | 39.6% |
| 2012 | 41.2% | 40.0% |
| 2013 | 40.8% | 39.8% |
| 2014 | 42.8% | 39.8% |
| 2015 | 41.8% | 39.8% |
| 2016 | 42.3% | 39.8% |
| 2017 | 41.5% | 39.4% |
| 2018 | 42.0% | 39.9% |
| 2019 | 42.2% | 39.5% |
| 2020 | 44.6% | 41.0% |
| 2021 | 43.1% | 40.7% |
| 2022 | 42.1% | 40.9% |
| 2023 | 41.5% | 40.2% |
| 2024 | 42.3% | 40.4% |
| 2025 | 42.2% | 40.6% |
The last decade looks different. Compare 2013–19, the last seven seasons of the 2010s, with 2021–25 (2020, the odd season above, sits out). The rate rose only about 0.3 points, too little to tell from luck: the true change most likely sits somewhere between −0.4 and +1.1 points. Shorter third downs added about 0.9 points. At the same distance, offenses converted about 0.5 points less often, also too small to tell from luck (most likely between −1.2 and +0.2). So at any given distance the rate did not rise. Counting only the seasons after the step, 2014–19, it slipped a little, by about 0.8 points. The shortening behind it came mostly after 2019, as the next section shows.
The 2014 step itself hasn’t gone away. At the same distances, offenses in 2021–25 still converted about 1.6 points more often than in 1999–2001, a little less than the 2-point gain across all of 2015–25. Two things could fake this split. Penalty flags rose over the same years, and a third-and-4 near the goal line is a different play from one at midfield. So we redid the split both ways. Leave out penalty first downs, and the same-distance gain from 1999–2001 to 2015–25 shrinks to about 1.4 points, with shorter distances supplying 39% of the rise. Grade each third down only against third downs from the same part of the field, and the same-distance gain is about 2.1. The table has every split, era by era. A range that stays above zero marks a part too big to be luck. In the 2010s-to-2020s row, the same-distance range crosses zero.
| Eras | Total change | From shorter distances | Same distance | Same distance, plays only | Same distance and field position |
|---|---|---|---|---|---|
| 1999–2001 to 2015–25 | +2.9 (+2.1 to +3.8) | +0.9 (+0.6 to +1.1) | +2.1 (+1.2 to +2.8) | +1.4 (+0.8 to +2.2) | +2.1 (+1.4 to +2.8) |
| 1999–2005 to 2021–25 | +2.3 (+1.5 to +3.0) | +1.1 (+0.8 to +1.3) | +1.2 (+0.6 to +1.9) | +0.8 (+0.2 to +1.4) | +1.3 (+0.6 to +1.9) |
| 2013–19 to 2021–25 | +0.3 (−0.4 to +1.1) | +0.9 (+0.7 to +1.1) | −0.5 (−1.2 to +0.2) | −0.4 (−1.1 to +0.2) | −0.5 (−1.2 to +0.1) |
Better early downs left shorter third downs
Third downs got shorter as first and second downs got better, and the change came late. The average third down needed 7.42 yards in 1999. It dipped once in the late 2000s: in 2006–09 it was 6.83–6.97 yards, and in 2008 29% of third downs needed 3 yards or less. Then it climbed back, and through the 2010s it barely moved. In 2020 it dropped to 6.78 yards and stayed down. Every season since has been shorter than any season of the 2010s, and about as short as 2006–09; 2020 was the shortest in our data. In 2025 it was 6.90. 2020 was also the hot season from earlier, with the highest same-distance rate in our data (43%, against 42% in 2014, 42% in 2019 and about 41% in 2021–25). That bump faded the next year; the shorter distances stayed.
Third downs got shorter after 2019
Share of third downs with 1–3 yards to go and with 7 or more, by season.
Show data
| Year | 7 or more yards | 1–3 yards |
|---|---|---|
| 1999 | 52.0% | 24.5% |
| 2000 | 50.3% | 26.4% |
| 2001 | 49.0% | 26.6% |
| 2002 | 48.8% | 26.8% |
| 2003 | 49.0% | 26.1% |
| 2004 | 48.8% | 26.5% |
| 2005 | 50.3% | 25.8% |
| 2006 | 47.9% | 26.6% |
| 2007 | 47.8% | 27.8% |
| 2008 | 46.8% | 29.0% |
| 2009 | 47.1% | 27.4% |
| 2010 | 49.3% | 26.9% |
| 2011 | 49.9% | 26.4% |
| 2012 | 48.2% | 27.3% |
| 2013 | 48.4% | 26.4% |
| 2014 | 48.7% | 27.1% |
| 2015 | 48.8% | 26.8% |
| 2016 | 48.5% | 27.0% |
| 2017 | 48.7% | 25.5% |
| 2018 | 48.5% | 27.3% |
| 2019 | 49.3% | 26.2% |
| 2020 | 44.6% | 29.4% |
| 2021 | 45.9% | 29.1% |
| 2022 | 45.7% | 29.4% |
| 2023 | 47.1% | 27.8% |
| 2024 | 46.0% | 28.8% |
| 2025 | 46.1% | 29.3% |
The shorter distances were made on first and second down. First-down plays averaged 5.1 yards in 1999 and 5.6 in 2025. A 6-yard gain on first-and-10 usually leaves the offense better placed to score than before the snap. A 1-yard run usually doesn’t. Count plays like the first as successes, and first-down plays succeeded 39% of the time then and 43% now. Second downs started shorter too, with 8.1 yards to go then and 7.7 now. A second-down play gained a first down 29% of the time in 1999 and 33% in 2025. When second down failed, it left third-and-3 or less 25% of the time in 1999 and 30% in 2025. It left third-and-7 or more 51% and 45% of the time.
Teams with better early downs also faced fewer third downs. In 1999, 50% of series reached one and 41% moved the chains on first or second down. In 2025, those shares were 48% and 46%. A team faced 13.9 third downs a game in 1999 and 12.9 in 2025. Team by team, the link is strong within a season: teams that face shorter third downs, and teams that do well on early downs, convert more of their third downs. The 2011 Saints’ third downs needed about 6.3 yards on average, and they converted 59%. The 2002 Cowboys’ needed about 8, and they converted 26%. Across 861 team-seasons, distance and third-down rate line up with a correlation of −0.57. On that scale 0 would mean distance tells you nothing, and −1 that shorter third downs always meant more conversions, in lockstep. Early-down success and third-down rate line up at 0.59. The downs before third changed more than third down itself. Over the same years, the air-yards story found early-down throws getting shorter while third-down air yards barely moved.
Third-and-long is as hard as it was in 1999
If the gains since the 2010s are all distance, the hardest third downs should look the same as ever. They do. Where offenses did get better, it was at the short and middle distances. On third-and-1, offenses converted 66% in 1999–2001 and 70% in 2021–25. On third-and-4 to 6, they converted 44% and 46%. On third-and-7 or more, they converted 27% then and 28% now, a gap too small to tell from luck. That rests on 10,328 third-and-longs in 1999–2001 and 16,532 in 2021–25. The true rate then most likely sat between 26.5% and 28.3%, and now between 27.3% and 28.7%. The two ranges overlap.
Short and middle third downs got easier. Long ones barely moved.
Share of third downs converted, 1999–2001 against 2021–25.
Show data
| 1999–2001 | 2021–25 | Change | |
|---|---|---|---|
| 3rd-and-1 | 66.0% | 69.8% | +3.7 pts |
| 3rd-and-2 | 55.1% | 58.4% | +3.3 pts |
| 3rd-and-3 | 50.9% | 52.0% | +1.2 pts |
| 3rd-and-4 to 6 | 43.9% | 46.5% | +2.5 pts |
| 3rd-and-7 to 10 | 33.7% | 34.5% | +0.8 pts |
| 3rd-and-11 or more | 17.7% | 17.8% | +0.1 pts |
Take third-and-7 or more when the quarterback dropped back to pass. Whether he then threw, took a sack or ran, offenses converted about 27% of the time in 1999–2005 and the same in 2021–25 (26.6% both times). (In between, in 2006–12, they converted a little more often, about 29%.) What changed was the mistakes. Interceptions fell from 3.7% of those dropbacks in 1999–2005 to 2.6% in 2021–25. Sacks rose a little, from 9.8% to 10.4%, which the sack story traces. The average third-and-long dropback still hurt the offense, just a little less. Analysts tally every play in points, by how much it moves the drive’s expected score up or down. An incompletion that brings on the punt team costs a little; an interception costs far more. On that tally the average third-and-long dropback cost about 0.14 points in 1999–2005 and about 0.11 in 2021–25. Scrambles are tagged reliably only from 2006, so for them we start with 2006–12. They went from 4% of dropbacks then to 6% in 2021–25, and converted 38% and 42% of the time.
Quarterbacks still throw short of the sticks, the first-down line. On third-and-7 or more, 54% of passes in 2006–12 fell short of it, and 57% in 2021–25. Those throws converted 19% and 18% of the time. Throws at or past the sticks converted 45% and 43%.
Our count runs a little above the NFL’s method, and the shape holds
Every rate in this story counts a defensive penalty that hands over a first down on third down as a conversion, so our numbers run a little above a plays-only count, the NFL’s method. The rise, the 2014 step and the shorter third downs look the same either way. The box below has the gap and the other limits.
Third down stopped being the last down
Offenses got better at third down once, and since 2020 they have faced shorter ones than in the 2010s. The bigger change came after third down failed. Take every drive series, a fresh set of downs starting on first down, that reached a third down. Of those series, 37.7% converted it in 1999–2001 (19,763 series) and 40.3% in 2021–25 (34,457). Add the series saved by a fourth-down conversion after a failed third, and the share that kept the drive alive went from 40.8% to 46.5%. That rise is about twice the rise on third down alone.
Third down stopped being the last down
Of drive series that reached a third down: the share that moved the chains on third, and the share that moved them on third or, after a failed third, on fourth.
Show data
| Year | On third or fourth | On third |
|---|---|---|
| 1999 | 39.8% | 36.8% |
| 2000 | 41.6% | 38.4% |
| 2001 | 40.8% | 37.8% |
| 2002 | 43.2% | 39.5% |
| 2003 | 41.3% | 38.0% |
| 2004 | 41.5% | 38.3% |
| 2005 | 42.0% | 38.7% |
| 2006 | 43.0% | 39.5% |
| 2007 | 44.1% | 40.2% |
| 2008 | 44.2% | 40.3% |
| 2009 | 43.3% | 39.2% |
| 2010 | 42.3% | 38.9% |
| 2011 | 41.6% | 38.8% |
| 2012 | 42.4% | 39.1% |
| 2013 | 42.3% | 39.0% |
| 2014 | 43.6% | 40.5% |
| 2015 | 43.1% | 39.6% |
| 2016 | 44.0% | 40.3% |
| 2017 | 42.5% | 39.2% |
| 2018 | 44.6% | 39.9% |
| 2019 | 44.2% | 39.8% |
| 2020 | 48.2% | 42.6% |
| 2021 | 47.2% | 41.2% |
| 2022 | 45.5% | 40.2% |
| 2023 | 45.4% | 39.7% |
| 2024 | 46.7% | 40.3% |
| 2025 | 47.5% | 40.4% |
After a failed third-and-1 to 3, teams went for it 19% of the time in 1999–2001 and 37% in 2021–25. Across all distances, fourth-down tries after a failed third converted 45% then and 54% now. Why coaches changed their minds is the fourth-down boom, told in its own story. Here it means a third down carries less weight than it did.
On third-and-short, teams went back to the run
The fourth-down shift may also explain a change on third-and-short. On third-and-1 to 3, the share of designed runs fell from 50% in 1999–2005 to 42% in 2013–19, then climbed back to 49% in 2021–25. Runs converted more often than dropbacks in every era. Inside the opponent 40, where going for it on fourth down is in play, the run share on third-and-1 or 2 traced the same V: 64%, 56%, 64%. That fits the idea that a safe third-down run costs less when fourth down is a real option. The V is in the play-by-play. The fourth-down reason is our guess.
| Era | Run share | Runs converted | Dropbacks converted | QB runs, share of 3rd-and-1 plays | QB runs converted |
|---|---|---|---|---|---|
| 1999–2005 | 49.8% | 64.7% | 51.6% | 11.7% (601) | 84% |
| 2013–19 | 41.6% | 66.2% | 54.4% | 10.6% (547) | 82% |
| 2021–25 | 49.1% | 67.3% | 53.5% | 19.3% (819) | 81% |
Quarterbacks took more of those runs themselves. On third-and-1, quarterback runs went from 12% of plays in 1999–2005 to 19% in 2021–25. They converted about 84% and 81% of the time. Those runs include the sneak and the tush push (a sneak with teammates shoving the quarterback from behind), which have their own story.
Some offenses really are better at third down
All of that describes the league as a whole. The next question is who converts third downs, and whether it lasts. Measure each team against what an average offense would convert at its distances, and two seasons sit above the rest: the 2011 New Orleans Saints and the 2006 Indianapolis Colts. The Saints, Sean Payton’s team with Drew Brees, converted 127 of 216 third downs (59%). That is about 15 more conversions per 100 third downs than an average offense would have made at the same distances. The Saints went 13–3. Theirs is the best raw rate in our data, and the season hardest to put down to luck, because the Saints kept that edge across 216 third downs. Counted in plain conversions per 100, the 2006 Colts were a hair further above average, +15.4 to the Saints’ +14.8. Brees threw for 5,476 yards that year, then an NFL record. On third-and-7 or more he converted 38 of 83 dropbacks (46%), the best of any quarterback-season with 80 or more.
Luck rarely makes a season like that. Imagine a league where every team were equally good at third down and only luck separated them. Even then, across 861 team-seasons, some team would stand out. The table below gives each team-season a beyond-luck score. An average offense scores 0. The score grows with the gap from average and with the number of third downs behind it, since the same edge is harder for luck to fake over more plays. In the all-luck league, the luckiest team would usually score near 3.2, and most likely between 2.7 and 3.8. The Saints score 4.6. The chance that luck alone pushes any team-season as far above average as the Saints is about one in 560. The next season the Saints still ran about 3 conversions per 100 above expectation. The Colts, Peyton Manning’s team under Tony Dungy, converted 109 of 189 (58%) and score 4.5, a one-in-350 shot for luck alone.
The bottom has its own outlier. Dave Campo’s 2002 Dallas Cowboys, with Chad Hutchinson and Quincy Carter at quarterback, converted 56 of 215 (26%) and went 5–11. Arizona’s 2012 season, the one with the 0-for-15 day, ranks second-worst on the beyond-luck score, level with the 2000 Browns, at 27%. Dallas sat further below average than Arizona, and that gap matters. Luck alone produces a season as bad as Dallas’s only about one time in 32. It produces one as bad as Arizona’s about one time in seven. Arizona’s year sits within luck’s reach.
| Season | Team | Coach | QB (most starts) | Record | Converted | Rate | Over avg. | Beyond luck (score) | Next season over avg. |
|---|---|---|---|---|---|---|---|---|---|
| 2011 | Saints | Sean Payton | Drew Brees | 13–3 | 127 of 216 | 58.8% | +14.8 | +4.6 | +2.8 |
| 2006 | Colts | Tony Dungy | Peyton Manning | 12–4 | 109 of 189 | 57.7% | +15.4 | +4.5 | +6.5 |
| 2004 | Vikings | Mike Tice | Daunte Culpepper | 8–8 | 106 of 198 | 53.5% | +12.9 | +3.9 | −3.7 |
| 2001 | St. Louis Rams | Mike Martz | Kurt Warner | 14–2 | 100 of 197 | 50.8% | +11.3 | +3.4 | +0.5 |
| 2023 | Cowboys | Mike McCarthy | Dak Prescott | 12–5 | 126 of 237 | 53.2% | +10.4 | +3.4 | −1.3 |
| 2002 | Cowboys | Dave Campo | Chad Hutchinson | 5–11 | 56 of 215 | 26.0% | −12.4 | −4.0 | −0.9 |
| 2012 | Cardinals | Ken Whisenhunt | John Skelton | 5–11 | 64 of 233 | 27.5% | −10.8 | −3.6 | −2.8 |
Continues below
| Season | Team | Coach | QB (most starts) | Record | Converted | Rate | Over avg. | Beyond luck (score) | Next season over avg. |
|---|---|---|---|---|---|---|---|---|---|
| 2000 | Browns | Chris Palmer | Doug Pederson | 3–13 | 58 of 209 | 27.8% | −11.4 | −3.6 | −3.4 |
| 2015 | St. Louis Rams | Jeff Fisher | Nick Foles | 7–9 | 59 of 206 | 28.6% | −11.6 | −3.5 | −7.4 |
| 2004 | Bears | Lovie Smith | Craig Krenzel / Chad Hutchinson (5 each) | 5–11 | 60 of 227 | 26.4% | −10.6 | −3.5 | −8.4 |
About half the gap between teams is real, and it carries over
The Saints and the Colts are the extremes. For most teams, the question is how much of a good or bad third-down year is real. In a typical season, teams land about 4.3 points either side of what an average offense would convert at their distances. A team plays roughly 219 third downs a season, few enough that luck alone moves its rate about 3.2 points either way. Take luck’s share out and about 2.9 points of real difference between teams remain. (Luck and skill don’t simply subtract. Show the work has the arithmetic.) Measured the way statisticians split a spread, skill accounts for about 46% of the differences between teams, about half. Luck accounts for the rest. Of 861 team-seasons, 127 sit so far from expectation that luck would put a given team there only one time in 20. Luck alone would produce about 43 such seasons, most likely between 33 and 53.
And the skill lasts. A team that sits above or below expectation on third down one season tends to stay on the same side the next, by less. The top fifth of team-seasons ran about 6 conversions per 100 above expectation and were still about 3 above the next year. The bottom fifth ran about 6 below and stayed about 2 below. Across 829 pairs of seasons, that carry-over shows up as a correlation of 0.38, most likely between 0.32 and 0.44. A 0 would mean a team’s edge vanishes the next year; a 1 would mean it repeats exactly. Red-zone touchdown rate, from our red-zone story, scores 0.21. So third down carries over nearly twice as strongly as the red zone. The table puts third down beside other things a team does.
| Measure (team against league) | Year-to-year link | Most likely between |
|---|---|---|
| Third-down rate, against distance | 0.38 | 0.32–0.44 |
Continues below
| Measure (team against league) | Year-to-year link | Most likely between |
|---|---|---|
| Third-down rate, raw | 0.43 | 0.37–0.48 |
| Early-down success rate | 0.42 | 0.36–0.48 |
| Average third-down distance | 0.34 | 0.27–0.40 |
| Third-and-7+ rate | 0.25 | 0.18–0.31 |
| Defense, third-down rate allowed | 0.15 | 0.08–0.21 |
| Red-zone touchdown rate | 0.21 | 0.15–0.28 |
Single seasons can still mislead. Dallas in 2023, under Mike McCarthy with Dak Prescott, sits fifth on the list above, a hair behind the 2001 St. Louis Rams, and fell to 1.3 per 100 below average the next year. Mike Tice’s 2004 Minnesota Vikings, with Daunte Culpepper, sit third and then fell to 3.7 per 100 below. Defenses carry over less: the third-down rate a defense allows one year says less about the next than an offense’s rate does (a link of 0.15, against the offense’s 0.38). Now try to guess a team’s third-down rate next season. Start from a blind guess, league average for everyone, and aim for a perfect call. This season’s third-down rate gets you about 18% of the way there; the rest is misses like Dallas’s fall after 2023. Early-down efficiency, the same points tally on first and second down, gets you about 15%. Use both and you get about 21% (829 pairs of seasons), so third down tells you something early downs miss.
Among quarterbacks, Philip Rivers’s third-and-long record is the hardest to call luck
If some teams really are better at third down, some of that skill should belong to the quarterback, and third-and-long is where it shows. The same test works for passers, and on third-and-7 or more the record hardest to put down to luck belongs to Philip Rivers. He converted 460 of 1,329 dropbacks (34.6%), about 5 more conversions per 100 than an average quarterback would have made at the same distances. He retired after the 2020 season, then came back in December 2025 to start three games for the Indianapolis Colts after Daniel Jones’s season-ending Achilles injury, and lost all three. Those dropbacks are in the count. A quarterback can run hot on third-and-long for a season. Rivers’s edge is spread over 1,329 dropbacks from 2004 to 2025, and luck almost never fakes an edge over that many. Per dropback, four quarterbacks were a little further above average over fewer plays, led by Jameis Winston (+5.9 per 100, against Rivers’s +4.9) and Patrick Mahomes (+5.8). Winston’s raw rate, 34.7%, is also higher than Rivers’s, and the highest of all belongs to Jimmy Garoppolo (34.9% of 307).
The twist is the other half of his game. On third-and-1 to 6, Rivers was a little below average. Three other quarterbacks near the top were good at both: Brees, Patrick Mahomes (223 of 644 on long) and Peyton Manning. The table has each one’s figures.
| Quarterback | Seasons | 3rd-and-7+ dropbacks | Converted | Rate | Over avg. (points) | Beyond luck (score) | Beyond luck on 3rd-and-1 to 6 |
|---|---|---|---|---|---|---|---|
| Philip Rivers | 2004–2025 | 1,329 | 460 | 34.6% | +4.9 | +4.0 | −0.6 |
| Drew Brees | 2001–2020 | 1,452 | 486 | 33.5% | +4.2 | +3.6 | +1.8 |
Continues below
| Quarterback | Seasons | 3rd-and-7+ dropbacks | Converted | Rate | Over avg. (points) | Beyond luck (score) | Beyond luck on 3rd-and-1 to 6 |
|---|---|---|---|---|---|---|---|
| Patrick Mahomes | 2017–2025 | 644 | 223 | 34.6% | +5.8 | +3.3 | +2.9 |
| Jameis Winston | 2015–2025 | 521 | 181 | 34.7% | +5.9 | +3.0 | −0.4 |
| Peyton Manning | 1999–2015 | 1,210 | 404 | 33.4% | +3.9 | +3.0 | +3.8 |
| Daunte Culpepper | 2000–2009 | 551 | 178 | 32.3% | +5.5 | +3.0 | +0.4 |
| Tim Couch | 1999–2003 | 332 | 65 | 19.6% | −7.0 | −2.9 | −1.9 |
| Sam Darnold | 2018–2025 | 503 | 118 | 23.5% | −6.0 | −3.0 | −0.2 |
| Alex Smith | 2005–2020 | 965 | 237 | 24.6% | −4.3 | −3.0 | −3.7 |
| Blaine Gabbert | 2011–2023 | 350 | 70 | 20.0% | −9.6 | −4.0 | −3.3 |
Over a whole career, quarterbacks really do differ. Of the 90 quarterbacks with 300 or more such dropbacks, 25 sit so far from average that luck would put a given passer there only one time in 20, 13 above and 12 below. Luck alone would give about 4 such careers, most likely between 1 and 8. It would put any career as far out as Rivers’s, or as low as Blaine Gabbert’s (70 of 350, 20%), only about a quarter of one percent of the time each. Single seasons are mostly noise: a quarterback’s third-and-long rate one year tells you little about his next. Across 512 pairs of seasons, the link is 0.22 on the same scale as the year-to-year table above.
Tom Brady’s last season shows how far one year can swing. In 2022, in Tampa Bay, he converted 12 of 91 dropbacks on third-and-7 or more (13%), the lowest against expectation of the 392 quarterback-seasons with 80 or more. Among that many seasons, luck alone produces one that low about one time in six. Brady may simply have run cold.
When third down fails now, the drive often goes on
Days like Arizona’s 0 for 15 have grown rare. In 1999–2005, 17 team-games in 3,530 went without a third-down conversion. In 2021–25, 6 in 2,718 did. A third-and-8 is about as hard to convert as it was in 1999. Offenses see fewer of them, convert more of the short ones, and face fewer third downs a game.
And a failed third down ends fewer drives. Of every 1,000 series that reached third down in 1999–2001, 592 failed to move the chains on third or fourth down. In 2021–25, 535 did. Teams converted about 27 more of those 1,000 on third down itself. But 57 fewer died.
Share this article
Team tools
Live Edgehalla tools on the same subject.
Show the work
- Tests written down
- Not in advance (exploratory)
- Results computed
The “Beyond luck (score)” columns in the team and quarterback tables are the gap from average in standard deviations of the luck alone would produce for that many third downs or dropbacks (a z-score). nflverse play-by-play, regular season 1999–2025 (188,443 third downs), with 2026 weeks 1–6 (the complete weeks in the file pulled Oct. 26, 2026) held out as a check. A third down is a 3rd-down run or pass (sacks and scrambles included); kneels, spikes, punts, field goals and two-point tries are out. A conversion is a first down on the play or an offensive touchdown. A no-play on which a defensive penalty gave a first down counts as a converted third down; any other no-play is replayed, and the replayed snap counts. nflverse codes some of those penalty first downs as no-plays and, from 2001, some on the play itself, so counting both is the only definition that doesn’t move with the coding. Plays only, the 2011 Saints are 119 of 208; the team’s site lists 118 of 208. The fixed-distance rate weights each season’s rate at each distance (1–20+ yards) by the 1999–2001 mix. Expected conversions for a team or quarterback use the league’s rate for that season and distance. Coaches, dates and starting quarterbacks come from the nflverse games file. Descriptive only: no causal claim.
| Test | Result | Verdict |
|---|---|---|
| One steady fixed-distance rate, 1999–2025 | χ² 80.5 on 26 df | Rejected |
More tests below
| Test | Result | Verdict |
|---|---|---|
| Break search: best single step in the fixed-distance rate | 2014: 40.0% before, 41.5% after; placebo p = 0.0005 (max statistic over 22 candidate years, 2,000 season-order shuffles) | Real |
| Range for the break year (game-level bootstrap) | 2010–2014 (90%); 2014 in 62% of 300 resamples | About 2014 |
| Step against a steady linear rise (fixed-distance rate) | within-segment χ² 31.8 on 25 df, p = 0.16 (after 2014 alone p = 0.10); straight line χ² 43.2 on 25 df, p = 0.01 | Step fits; line doesn’t |
| Break search, raw rate | 2014: 40.5% → 42.4%; placebo p = 0.0005; bootstrap 90% 2006–2014; within-segment p = 0.003 (both segments; after 2014 p = 0.03, mainly 2020) | Same year |
Continues below
| Test | Result | Verdict |
|---|---|---|
| Break search, plays only (no penalty first downs) | 2014: 38.0% → 38.9%; placebo p = 0.002; bootstrap 90% 2004–2014, 2014 in 46% of resamples | Same year, wider range |
| Break search, distance and field zone held fixed | 2014: 40.0% → 41.5%; placebo p = 0.001; bootstrap 90% 2010–2014 | Same year |
| Pooled rate, 1999–2001 vs 2015–25 | 39.4% (n = 20,484) vs 42.3% (n = 76,998); 2015–25 seasons 41.7–42.9% (t-interval) | Real rise |
| Split of the change, 1999–2001 to 2015–25 (percentage points, 95% ranges from resampling games; Kitagawa) | total +2.9 (+2.1 to +3.8); from shorter distances +0.9 (+0.6 to +1.1); same distance +2.1 (+1.2 to +2.8); plays only (no penalty first downs), same distance +1.4 (+0.8 to +2.2); same distance and field zone +2.1 (+1.4 to +2.8) | Both parts real |
Continues below
| Test | Result | Verdict |
|---|---|---|
| Split of the change, 1999–2005 to 2021–25 (percentage points, 95% ranges from resampling games; Kitagawa) | total +2.3 (+1.5 to +3.0); from shorter distances +1.1 (+0.8 to +1.3); same distance +1.2 (+0.6 to +1.9); plays only (no penalty first downs), same distance +0.8 (+0.2 to +1.4); same distance and field zone +1.3 (+0.6 to +1.9) | Both parts real |
| Split of the change, 2013–19 to 2021–25 (percentage points, 95% ranges from resampling games; Kitagawa) | total +0.3 (−0.4 to +1.1); from shorter distances +0.9 (+0.7 to +1.1); same distance −0.5 (−1.2 to +0.2); plays only (no penalty first downs), same distance −0.4 (−1.1 to +0.2); same distance and field zone −0.5 (−1.2 to +0.1) | Distance only |
| Holdout: 2026 weeks 1–6 vs same weeks 2015–25 | 42.5% (n = 2,471) vs 42.6% (n = 26,956); fixed-distance 2026 41.6% | On the plateau |
Continues below
| Test | Result | Verdict |
|---|---|---|
| 2020 | 44.6% raw, 42.8% fixed-distance; mean distance 6.78 yards | One-season outlier, no cause assigned |
| Third-and-7+: 1999–2001 vs 2021–25 | 27.4% (95% range 26.5–28.3%, 10,328 plays) vs 28.0% (27.3–28.7%, 16,532 plays) | Flat |
| Team third-down rate against expectation, year to year | r = 0.38 (0.32–0.44, n = 829); without 2020 pairs 0.38; split-half 0.32, full-season 0.48 | Repeats |
| Signal share of single-season spread | observed SD 4.3, luck SD 3.2 (about 219 third downs a team-season), true SD 2.9 points; the parts add as squares, 4.3² ≈ 3.2² + 2.9²; signal share = true² / observed² = 46% (a share of the variance, not of the 4.3-point spread) | About half skill |
Continues below
| Test | Result | Verdict |
|---|---|---|
| Team-seasons beyond ±1.96 vs luck-only simulation | 127 of 861 vs 43 (90% 33–53) | Real skill |
| Most extreme of 861 team-seasons in a luck-only league (z) | best: median 3.16 (90% 2.72–3.80); worst: −3.15 (−3.88 to −2.70); any as far as 2011 NO 4.60: 0.2%; 2006 IND 4.51: 0.3%; 2002 DAL −3.96: 3.1%; 2012 ARI −3.56: 14.8% (exact normal tail) | Saints, Colts, Cowboys beyond luck; Cardinals within |
| Next-season third-down rate: R² from this season’s rate, early-down EPA, both | 18.1% / 15.4% / 21.2% (n = 829) | Third down adds its own information |
| Quarterback careers (300+ third-and-7+ dropbacks) beyond ±1.96 | 25 of 90 vs 4 by luck (90% 1–8); QB-season r 0.22 (n = 512) | Signal over a career; names picked after the fact |
Continues below
| Test | Result | Verdict |
|---|---|---|
| Luck-only extremes, quarterbacks | career best z median 2.44 (90% 1.86–3.25); Rivers 4.02: 0.26%; Gabbert −4.03: 0.25%; season worst of 392: Brady 2022 (12 of 91, 13.2%) −3.31: 16.8% | Rivers and Gabbert beyond luck; Brady 2022 within |
| Longest third down converted | 3rd-and-37, Oct. 31, 1999, Vikings at Broncos: 53-yard run | Scene, not evidence |
| Most conversions in a game (15) | 8 team-games, Nov. 24, 2002 to Sept. 12, 2021 | Ties kept |
Frequently asked questions
What is a good third down conversion rate in the NFL?
The league converted 42.2% of third downs in 2021–25, counting first downs from defensive penalties, with seasons from 41.5% to 43.1%. A plays-only count, the NFL’s method, runs about 2.4 points lower. The top fifth of teams since 1999 ran about 6 points above what an average offense would convert at their distances.
Which NFL team had the best third down percentage?
In play-by-play data since 1999, the 2011 New Orleans Saints: 127 of 216 by our count (58.8%), or 118 of 208 on the team’s own play-only count. The 2006 Indianapolis Colts (57.7%) are next. Both sit beyond what luck would produce.
Do NFL teams convert third down more than they used to?
Yes, by about 2.9 points since 1999–2001. Most of that came from converting the same distances better, and that gain was complete by about 2014. After that the rate at each distance slipped slightly. From 2020, third downs were shorter than in the 2010s, which kept the overall rate level.
Is third down conversion rate predictive?
More than most team stats. A team’s third-down rate against expectation correlates 0.38 with the next season’s, against 0.21 for red-zone touchdown rate. About half of a single season’s spread is real skill, and this season’s third-down rate predicts next season’s about as well as early-down efficiency does.
Has third and long gotten easier in the NFL?
No. Offenses converted 27.4% of third-and-7 or more in 1999–2001 and 28.0% in 2021–25, a gap inside noise. Interceptions fell, but the conversion rate didn’t move.
More in The long arc
- Verdict: Data story
Whatever Happened to the No-Huddle?
- Verdict: Data story
NFL Coaches Throw About Half as Many Red Flags as They Did in 2007
- Verdict: Data story
A Quarter Fewer NFL Deep Throws Go Down the Middle
- Verdict: Data story
NFL Red-Zone Touchdowns Rose in Two Jumps, Then Stopped
- Verdict: Data story
About Seven NFL Head Coaches Lose Their Jobs Every Year, and Always Have
- Verdict: Data story
Good NFL Teams Stay Good for Less Time Than They Used To
Sources and method
- nflverse play-by-play and players (CC-BY 4.0)
- nflverse games file (nflverse/nfldata): results, dates, head coaches and starting quarterbacks, 1999–2025
- Pro Football Talk (NBC Sports), “Expect more defensive holding, illegal contact flags this year”, Aug. 1, 2014
- nflpenalties.com, 2014 point-of-emphasis penalties (defensive holding after week 1)
- New Orleans Saints, 2011 regular-season team statistics (neworleanssaints.com)
- 2011 New Orleans Saints season (Wikipedia): 13–3 under Sean Payton; Brees’s 5,476 passing yards
- WRTV Indianapolis, “Philip Rivers to start Sunday for Colts against Seahawks”, Dec. 13, 2025
- Michael Marot (AP), “Philip Rivers is happy he got 3 bonus games, even if he’s made his last start for the Colts”, Dec. 28, 2025 (via WSLS)
Research and analysis, not betting advice. Past results do not guarantee future returns; bet only what you can afford to lose.