← All articles

Strength of schedule is a forecast, not a fact

Every spring, as soon as the NFL releases its schedule, the same list starts circulating: all thirty-two teams ranked from hardest slate to easiest, usually to two or three decimal places. The list gets quoted in win-total previews, fantasy draft guides, and season predictions, and it carries the quiet authority of arithmetic. It deserves much less. A preseason strength-of-schedule figure is not a description of anything that has happened. It is a forecast, built from one of the weakest predictors available, and even the in-season version that replaces it is noisier and more circular than its precision suggests.

Where the preseason number comes from

The standard method is simple enough to do on a napkin. Take each opponent on a team’s schedule, look up that opponent’s win-loss record from the previous season, and combine the records into a single winning percentage. Teams facing opponents who went a combined well over .500 last year have a “hard” schedule; teams facing a combined losing record have an “easy” one. The figure is then treated as a property of the coming season, as though the opponents were fixed objects whose difficulty had already been measured.

Notice what the calculation actually assumes: that last year’s record is a good estimate of this year’s quality, that every game against a given opponent is equally hard, and that a team’s record captured its true strength in the first place. Each of those assumptions is shaky on its own. Stacked together, they produce a number that looks precise and mostly is not.

Last year’s records regress hard

An NFL season is seventeen games. That is a tiny sample by the standards of any other major league, and a large share of those games are decided by one score or less. A team that wins most of its close games will finish with a record well above its underlying quality; a team that loses them will finish well below. Close-game outcomes are only weakly repeatable from one year to the next, so the record a team posted last season already contains a substantial slice of luck before anything about the roster changes.

Luck does not carry over, which is why records regress toward the middle. The team that went 13–4 on a string of late-game breaks is, in expectation, a good team that will win fewer games next year. The team that went 4–13 while losing a pile of one-score games is probably better than its record. A preseason schedule metric built from raw records treats both of those teams at face value, which means it systematically overstates how hard it is to play last year’s overachievers and how easy it is to play last year’s unlucky teams. The spread between the “hardest” and “easiest” schedules is inflated by noise from the start.

Rosters do not carry over

Even a perfectly measured record from last season would describe a different team. NFL rosters turn over heavily every offseason through free agency, the draft, trades, retirements, and cuts. Coaching staffs change, and with them whole offensive and defensive systems. The single most important variable — who plays quarterback — can flip entirely between February and September. A team that lost its starting quarterback in Week 3 last year and limped to a losing record is not the same opponent once he returns healthy. A team that rode a veteran passer to the playoffs and then lost him to retirement is not either.

The preseason metric is blind to all of it. It gives full weight to an opponent’s old record regardless of whether the players who produced that record are still in the building. That is the core reason preseason strength of schedule correlates only loosely with how difficult a schedule turns out to be once the games are played. The number is measuring the ghost of last year’s league.

Mid-season SOS is circular and noisy

Once games are played, most outlets switch to current-season opponent records, which sounds like an improvement and partly is. But it introduces a different problem: circularity. A team’s opponents have records partly because of the games they played against that team. Beat a string of opponents and you have made their records worse, which makes your own schedule look easier, which makes your wins look less impressive. Lose to them and the reverse happens. The measure contaminates itself, and the effect is largest early in the year, when each game is a big share of every opponent’s sample.

The samples are small throughout. Halfway through the season an opponent has played eight or nine games, and its record still carries most of the noise described above. Injuries compound it. An opponent’s season-long record blends games with its starting quarterback and games without him, games with a healthy offensive line and games with backups, into a single figure that describes none of those versions accurately.

Timing matters too, and win-loss SOS ignores it completely. Facing a contender in Week 2, before a key injury, is a different assignment than facing the same team in Week 15 with its backup quarterback starting. Catching an opponent on a short week, at the end of a long road trip, or after it has clinched and is resting starters can make a nominally hard game soft. Two teams can share an identical “opponent winning percentage” while one of them drew every strong opponent at full strength and the other drew the same opponents at their weakest. The metric cannot tell them apart.

Better ways to measure who you played

None of this means schedules do not matter. They do, and adjusting for them is worthwhile. The fix is to adjust with better inputs. The first is point differential instead of win-loss record. Margin of victory carries more information per game than a binary win or loss, is less sensitive to the coin flips of one-score finishes, and predicts future results more reliably than record does over the same number of games. An opponent strength built on points scored and allowed is a meaningful upgrade with no extra complexity.

The second is opponent-adjusted efficiency. Rather than asking whether an opponent won, ask how efficiently it moved the ball and prevented the other side from doing so, play by play, and then adjust each team’s numbers for the quality of the units it faced. Rating systems built this way solve for every team’s strength simultaneously, which removes most of the circularity: an opponent’s rating is informed by all of its games, not just its record against you. These systems can also be split by unit, so a team facing a run of strong pass defenses can be distinguished from one facing strong run defenses, which matters for both game planning and individual player projection.

The third is to stop presenting schedule difficulty as a single point. A preseason projection should be a range: given the uncertainty about each opponent’s quarterback situation, roster turnover, and regression, how wide is the plausible band of difficulty, and how much of the gap between teams survives that uncertainty? Run that exercise honestly and most of the league clusters in the middle, with only a handful of schedules distinguishable from average with any confidence. That is a less exciting list to publish. It is also a more accurate one.

The honest read

Strength of schedule answers a real question — how hard was, or will be, the path this team faced — but the popular versions answer it badly. The preseason figure leans on last year’s records, which are inflated by luck and describe rosters that no longer exist. The in-season figure is circular, built on small samples, and blind to injuries, quarterback changes, and the timing of each matchup. Treat any schedule ranking as a forecast with wide error bars, prefer measures built on point differential and opponent-adjusted efficiency, and be skeptical of anyone who explains a win-total projection by pointing to a schedule ranking computed to three decimal places.