Defense Moves More Than Offense. Measure It Better and It Does Not.

You take over a struggling team. Do you fix the offense or the defense? It is the most basic question in running a club, and it has a clean answer available — which is exactly why it is worth checking twice.

Start with something settled. A point of offense and a point of defense are worth identically the same in win probability. Not roughly. Identically, always, in every league. Every game where one team scores 24 is the game where the other allows 24, so the two curves are mirror images and their slopes match everywhere. There is nothing to find there.

So the only question left is which side actually moves.

The answer, in points

We took 797 NFL clubs with three straight seasons, scored every offense and defense against its own league year, and asked how often a unit jumped a full standard deviation.

Defense, 21.1%. Offense, 17.1%. A four-point edge, and when we resampled the clubs ten thousand times the interval ran from +0.5 to +7.5 — never touching zero. Fix the defense. It is the one that budges.

That is a clean finding and we nearly published it.

Then we changed the ruler

Our editor pushed back with one line: points are tied too directly to the outcome, so use yards per play. He is right, and here is why it matters. A defense that is losing by three scores faces an offense running the clock. A defense whose offense keeps turning the ball over gets worse field position all afternoon. None of that is defending; all of it lands in points allowed.

Yards per play is not perfect, but it is closer to what a unit did rather than what happened to it. Same 797 clubs, same method, different ruler:

The edge falls from +4.0 to +1.5, and the interval now runs from −2.3 to +5.3. It crosses zero. On the better measure we cannot tell the two sides apart.

Notice what did not change: defense still swings around more, moving a full standard deviation either way 37% of the time against offense's 33%. But it swings both ways. It gains no more often than it loses. That is volatility, not fixability, and the two look identical if you only count the good half.

Why points lied

There is a second clue in the data. Measure how consistent a unit is from week to week and you can work out how reliably a season pins it down. Offense comes out at 0.75. Defense at 0.58.

A number measured worse bounces more. It looks changeable because it is blurry, not because it is soft. Correct for that and the year-over-year stickiness of the two sides is 0.56 and 0.60 — the wrong way round from the raw figures, and close enough to call level.

College is a different sport

We ran the same thing on 2,230 college clubs since 2005. Everything sticks harder there: corrected stickiness of 0.71 on offense and 0.82 on defense, against 0.56 and 0.60 in the NFL. A college defense is the least movable thing in either sport.

That fits what college is. There is no draft handing the worst team the best player, and the gap between the top and bottom is enormous — clubs are spread nearly twice as far apart on offense as NFL clubs are. Turning a program around is genuinely harder, and the numbers say so before anyone argues about it.

What to take home

When a finding depends on which number you counted, the number is the finding. We had a result with a clean interval and we would have printed it. The thing that killed it was not more data. It was one question: is this a fact about football, or a fact about points?

Ask that of any statistic that surprises you. Especially one that flatters a decision somebody is about to make with real money.

We checked the schedule too

The obvious objection is that a unit did not improve, it just drew easier opponents. So we built the adjustment — solve every offense and every defense at once, so each team is credited against the units it actually faced — and ran the whole thing again.

It barely moves anybody. Adjusted ratings correlate with raw ones at 0.98 on offense and 0.95 on defense; the typical club shifts about a fifth of a standard deviation. And the gap we were testing does not budge at all: +1.5 points either way, interval still crossing zero.

That is a result about the NFL, not about schedule adjustment. The league already balances its own fixtures — a division played twice, the rest rotated — so there is little schedule left to remove. In college, where a team can miss most of its conference and never leave the state, we would expect the same adjustment to matter enormously. We have not done that one yet, and we are not going to imply otherwise.

What this still cannot do. We treat a season as one number, which hides a club that fixed itself in November. The college figures here use raw scores against wildly unequal opposition, so read them as the coarser half of the piece. And none of this says how a unit got better — only that it did.

Run any of it yourself; the script is public. More at thesportspage.net.