The Sports PageThe Booth Shows You Twelve Numbers. Eight of Them Mean Nothing.
A game hands you a dozen stats. We tested all of them against 5,278 games. Four tell you something. The rest just tell you the score again.
Can these numbers tell me if a team is good?
Four of them can. Eight cannot.
Here is the surprising part. You would expect the bad ones to look bad. They do not. Every single stat here looks useful at first; third down percentage, first downs and turnover margin all track winning teams closely. They fall apart only when you check them against something you already knew — the score.
That is the whole answer. Here is how we know.
How did you test them?
We took ten years of games. For each team we measured every stat over the first half of its season.
Then we asked one thing. Did it predict the second half?
The second half is the test. Those games came later. The stat never saw them.
This matters more than it sounds. Test a number on the games you built it from and it will look brilliant. It will also be worthless. We learned that the hard way, and we wrote about it.
Which four are worth watching?
| Stat | What it does |
|---|---|
| Yards per play | The best of them |
| Explosive plays (20 yards or more) | Nearly as good |
| Completion percentage | Real, and small |
| Total yards | Real, and small |
Notice what they have in common. Each one is a rate. Yards for every play. Big plays out of all plays. Catches out of throws.
And the eight that do not work?
| Stat | Worth, once you know the score |
|---|---|
| Passing yards | +0.04 |
| Sacks | +0.03 |
| Third down percentage | +0.03 |
| First downs | +0.03 |
| Rushing yards | +0.03 |
| Red zone touchdown rate | +0.00 |
| Time of possession | −0.00 |
| Turnover margin | −0.02 |
Zero means zero. These are counts, not rates. A team gets more of them by having the ball more, and it has the ball more because it is winning.
Look at the two dots on each line. The dark dot is how good the stat looks on its own. The light dot is what survives once you know the score.
Eight of them collapse into the shaded strip. That strip is zero.
Why does third down percentage fail? It is on every screen.
Because it is telling you who was ahead.
A team with a lead faces short third downs. It ran twice and gained six yards. Now it needs four.
A team behind faces long ones. It threw twice, both fell incomplete. Now it needs ten.
The good rate did not build the lead; the lead built the good rate.
If the stat is a count, it is describing the game you just watched. If it is a rate, it might be telling you about the next one.
The ProfessorBut every coach says turnovers decide games.
They do decide games; they just do not predict them.
Those are two different claims and they get mixed up constantly.
A fumble can lose you a Sunday. But a ball hits the ground and bounces. Roughly half the time the offense gets it back. No coach changes the shape of a football.
So a team that won the turnover battle all year is barely more likely to win it again.
What about time of possession?
Worth nothing. Less than nothing, in our numbers.
Hold the ball for 36 minutes and you probably led. Leading teams run. The clock keeps moving. Trailing teams throw, and the clock stops.
The holding did not cause the win; the win caused the holding.
So what should I do on Sunday?
Work out yards per play in your head. It takes five seconds.
Take a team’s total yards. Divide by its plays. Do it for both teams.
Six yards a play against four is a real gap. Both graphics are on the screen already. Nobody does the division for you.
What to take home
One question beats all of this. When a number comes up, ask: is it a count, or is it a rate?
Counts describe what already happened; rates might tell you what happens next.
You will be right most of the time, and you did not need our table to do it.
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
5,278 team-games, 2016–2025, from nflverse play-by-play. Each stat is a margin — own minus opponent — averaged over weeks 1 to 8, then correlated with win rate from week 9 on. 320 team-seasons with at least six games either side.
The second column is a partial correlation holding point margin fixed. On its own, every stat here looks useful: third down percentage +0.280, first downs +0.290, turnover margin +0.245, possession +0.216. Holding the score fixed they fall to +0.033, +0.031, −0.016 and −0.003. The four that survive are yards per play +0.081, explosive plays +0.070, completion percentage +0.068 and total yards +0.061.
Penalties are credited to whoever committed them rather than assumed to be the offense. Red zone trips are counted once per drive, not once per play inside the twenty, which would reward stalling there.
Three limits. This predicts a team’s remaining SEASON, not a single game; third down percentage plainly matters in the game you are watching. "Adds nothing" means adds nothing beyond the score — if you somehow knew the third down rate and not the result, you would still know something. And one oddity we are not explaining: penalty yards runs slightly the wrong way, so committing fewer penalties tracks a marginally worse second half. It is small, we have logged it, and we would rather leave it unexplained than invent a reason.
Tools: scripts/build_nfl_broadcast_stats.py and scripts/rank_broadcast_stats.py.
