Going for it on fourth down: how the math finally won
For most of NFL history, fourth down was a binary cultural decision, not a mathematical one. Short yardage near midfield: punt. Long yardage anywhere short of the goal line: punt. The end-of-game exception was the only place coaches went for it without a sense that they were doing something risky enough to require an excuse afterward. That this was the consensus answer was strange even at the time, because the math had been quietly screaming the opposite answer since at least the 1970s. The math is still screaming it. It just took fifty years for coaches to start listening.
The interesting question isn't whether the analytics are right — they are, robustly, on every dataset anyone has run them on. The interesting question is why football resisted the answer for so long, and why it eventually stopped.
What the math says
The basic framework was sketched out by economist David Romer in a 2002 paper called "It's Fourth Down and What Does the Bellman Equation Say?" Romer modeled fourth-down decisions in expected- points terms. For every field position and every distance to a first down, he calculated the expected points from going for it (some probability of a conversion, leading to a drive that has its own expected point value, weighed against the cost of failing and handing the ball over) versus the expected points from punting (the field-position swing).
The result, summarized loosely: NFL teams should go for it on fourth down dramatically more often than they did. Not just on fourth-and-one near midfield — that was barely controversial — but in lots of situations that the conventional wisdom said were obvious punts. Fourth-and-three from your own forty? Go. Fourth-and-five at midfield? Go. Fourth-and-goal from the four, down by two scores in the third? Definitely go.
Romer's paper estimated that teams were leaving roughly half a win per season on the table by punting too much. Subsequent models, run on more recent data with better play-by-play resolution, have all landed in the same neighborhood: punting on shorter fourth downs in opponent territory is bad, kicking field goals from inside the ten on fourth-and-short is bad, and the one place coaches were roughly correctly aggressive — fourth-and-goal from the one — was the place where almost nobody disagreed in the first place.
Why coaches resisted
Several things all pulling in the same direction. The first is loss aversion in the most literal sense. A failed fourth-down attempt is a vivid, scrutinizable, on-camera mistake that the coach owns alone. A punt that hands the opponent the ball at their forty is a non-decision; it can't be questioned afterward because no one was asked to choose. Coaches were rationally optimizing for a non-rational thing — the asymmetry of public blame — at the cost of expected points.
The second is that coaches were trained inside a system that treated fourth down as a moral question rather than a probabilistic one. "We trust our defense" was the standard phrase, and it implied that going for it was a vote of no confidence. The math says it isn't — the question is which choice produces more points in expectation, not which choice expresses more faith in any unit.
The third is that the math wasn't legible to most coaches when it was first published. Romer's paper was a Bellman equation in an economics journal. It took analytics translators inside front offices, working through the same numbers in football vocabulary, before anyone could communicate the result up the chain to a head coach in a way that didn't sound like a graduate seminar.
How it changed
The shift didn't come from any one team's revelation. It came from a slow accumulation of evidence and the rise of decision-support software inside team operations. By the late 2010s, multiple front offices had built or bought "fourth-down models" — software that produced a real-time recommendation on every fourth-down play, weighted by score, time, and field position. Some teams had a coordinator with a tablet on the sideline whose only job was to relay the model's call to the head coach.
The shift was generational rather than ideological. The coaches who adopted the math first were generally the ones with the weakest cultural ties to the older football establishment — the ones who came up through college programs that had already gone analytical, the ones who were comfortable saying out loud that they ran a model on this. The holdouts, in turn, were increasingly losing late-game decisions to opponents whose tablets said go.
By the early 2020s the league-wide go-for-it rate on fourth-and- short had roughly doubled from where it had been a decade earlier. Punting on fourth-and-one inside the opponent's forty had moved from default to outlier. The math wasn't fully won — coaches were and are still less aggressive than the models — but the direction was clear, and the direction was permanent.
What the math still doesn't capture
It's worth being honest about what the analytics don't see. Most public fourth-down models assume average offense and average defense; a team with a historically bad short-yardage line might rationally punt in a spot where the league-average team should go. The models assume average kicker; a team with a 65-yard-leg kicker should kick field goals the average team shouldn't. The models don't see fatigue, weather past a crude threshold, or the specific fact that this offense has not converted a third-down all afternoon. Real coaching decisions live inside a fog the models can only partly clear.
And there's a real game-theoretic point that the simple models miss: if coaches go for it 100% of the time on fourth-and-one, defenses adjust, and the conversion rate falls. The current equilibrium is somewhere between the old culture and the math's recommendation, and it should be — the optimum isn't the model's recommendation in isolation, it's the model's recommendation accounting for the opponent's response.
The honest verdict
Fourth-down decision-making is one of the cleanest cases in sports of analytics being right and the establishment being wrong for decades. It is also a cautionary tale about how long it takes for an unambiguously correct answer to become operational behavior when the cultural cost of being wrong is asymmetric. The math didn't change. The data didn't change. What changed was the career risk of ignoring it.
It's tempting to read this as a story about how analytics won. The more accurate read is that fourth-down decision-making is the easiest decision in football for analytics to win. The variables are bounded, the outcome is observable, the time horizon is short, and the payoff is immediate. Most decisions in sports are not like that. The harder analytical questions — how to evaluate a rookie quarterback after eight starts, how to set a salary-cap structure five years out, how to balance load management against playoff seeding — don't have the same mathematical clarity. Fourth down was the test case. The rest of the league is now in the harder cases.