The Sports PageTurnovers Win Games. They Do Not Win Seasons.
A season’s turnover margin explains 3% of the next season’s. How a team moves the ball explains 15%. Turnovers feel like character and behave like weather.
The question: your team lost the turnover battle again. Is that who they are?
Almost certainly not. A season’s turnover margin explains 3% of the next season’s. How a team moves the ball explains 15%.
Why that is surprising. Turnovers feel like character — discipline, ball security, coaching. They behave like weather.
What you get: a way to know which of this year’s numbers will still be true next year.
The test
Take every team season from 2016 to 2025. Measure something. Then measure the same thing for the same team the following season, and see whether the two move together.
That is the plainest question you can ask about a statistic: does it describe the team, or does it describe the year?
Yards-per-play margin carries 15.4% into the next season. Win rate carries 13.6%. Turnover margin carries 3.0%.
Three percent. Across 288 team-seasons.
What three percent means
It does not mean turnovers are unimportant. They decide games; a fumble at the wrong moment has ended seasons.
It means they do not repeat. A team that won the turnover battle all year is barely more likely than anyone else to win it again.
Consider what a turnover actually is. A ball hits the ground and bounces. Twenty-two men dive. Roughly half the time the offense gets it back, and no amount of coaching changes the shape of a football.
Interceptions have a little more skill in them. Recovery does not, and recovery is most of it.
How a team moves the ball repeats. How the ball bounces does not.
The ProfessorWhat to do with this in September
Your team is 1–1 and minus-four in turnovers. Everyone is talking about discipline.
Hold onto the three percent. Most of that margin is going to come back on its own, and the coaching change people are demanding will get credit for the regression.
Now flip it. A team at plus-five after two weeks is not a disciplined team; it is a lucky one, and it is going to look worse in November while playing exactly as well.
This is the same shape as Detroit’s one-run record in baseball and the same shape as a week-one blowout in college football. We keep finding it because it keeps being true: the extreme thing regresses, and the thing that regresses is the thing everybody built a story on.
What to take home: separate the team from the year
Before you believe a statistic about your team, ask one question: would this still be true next season?
For yards per play, mostly yes. For turnover margin, mostly no.
You do not need our table to do this. You need the habit of asking — and a healthy suspicion of any number that arrives already wrapped in an explanation.
Baseball is ranked by championship leverage: we simulate the rest of the season 60,000 times, then split those seasons by who won each of today's games. College football cannot be simulated that way, so it is ranked by how close to a coin flip the ratings make it — a lopsided game teaches you nothing. Tomorrow this block reports what happened to today's pick.
Notes & sources
Built from nflverse play-by-play, 2016 through 2025, aggregated to one row per team per game by scripts/build_nfl_history.py: 5,278 team-games with scrimmage plays, yards, drive-level possession, turnovers, points and result. Analysis in scripts/nfl_predictive.py.
Year-over-year persistence across 288 consecutive team-season pairs, 2016–2025. Correlation of a team’s season average with the same team’s next season: yards-per-play margin r = +0.392 (t = +7.20), win rate r = +0.369 (t = +6.71), turnover margin r = +0.175 (t = +3.00). The percentages quoted are r-squared: 15.4%, 13.6% and 3.0%.
Turnover margin is takeaways minus giveaways, counting interceptions thrown and fumbles lost. Fumbles FORCED are excluded from the giveaway side deliberately — a forced fumble recovered by the offense is not a turnover, and lumping the two together is what makes the statistic look more repeatable than it is.
Three limits. Persistence across seasons is a low bar: rosters, coaches and schedules all change, which is why even the best figure here is only 15%. A low year-to-year correlation does not prove there is no skill in ball security — it shows that whatever skill exists is swamped by variance at a 17-game sample. And r-squared measures only the straight-line relationship; a team that genuinely fixed something would be invisible inside an average of 288.
