WaltEvery spring, the same argument runs on repeat. A quarterback throws for 4,300 yards and 32 touchdowns, and half the league’s fan bases decide he has arrived. Another one cuts his interception rate in half and gets labeled a changed player. Then September comes, and the numbers move again, sometimes in the opposite direction, and nobody can agree on whether the previous year meant anything at all.
The problem is that not all quarterback statistics carry the same amount of information about the future. Some repeat almost mechanically from one season to the next. Others behave close to the statistical noise. Sorting the two apart is not a matter of opinion, and it does not require anything more complicated than a measure most people have already learned in a first statistics course – the correlation coefficient.
Year-over-year stability, and a small example you can rerun
The standard test is simple. Take every quarterback with a large enough sample in Year N and in Year N+1, pair the two values of a given statistic, and compute the strength of the linear association between them. A high value means the statistic tends to persist. A value near zero means last season told you almost nothing about the next one.
Here is a scaled-down version with six quarterbacks. First, interception rate (percent of passes intercepted):
- Year N: 1.5, 2.8, 1.9, 3.2, 2.1, 2.5
- Year N+1: 2.0, 2.6, 2.2, 2.4, 2.9, 1.7
By computing the correlation between these two datasets, you get roughly 0.21 — meaning a weak association. The quarterback with the lowest interception rate in Year N is nowhere near the lowest the following year. Now do the same for sack rate:
- Year N: 4.2, 8.5, 5.1, 7.8, 6.0, 9.1
- Year N+1: 4.6, 8.1, 5.5, 7.2, 6.4, 8.8
That pair returns about 0.99, revealing a strong positive linear correlation. The order barely changes. Same six passers, same two seasons, two completely different pictures — and that gap is the entire point.
Run the full test across the real box score, and the same hierarchy appears. Sack rate and pressure-to-sack rate are among the stickiest quarterback numbers in football, which is why the analytics community stopped treating sacks as a purely offensive-line outcome; they track tendencies that show up on tape, like how quickly a passer resets his eyes or whether he drifts backward under duress.
None of this says interceptions do not matter. They obviously decide games. It says that a single season’s interception rate is a weak predictor of the next season’s interception rate, which is an entirely different claim.
Turning stability into a projection
When a statistic has low year-over-year stability, the correct forecast is not the player’s own number. It is something pulled back toward the league’s central value for that statistic.
You can compute that baseline from the same example. Take the six-year N+1 interception rates — 2.0, 2.6, 2.2, 2.4, 2.9, 1.7 — put them into a mean calculator, and you get exactly 2.3. That is the number every quarterback is drifting around.
Now, suppose a passer posted a shiny 1.5 percent interception rate, and you want to project next season. Because the statistic is unstable, you do not carry 1.5 forward. You take a blended estimate of the player’s mark and the league baseline, with weights determined by the statistic’s stability. Interception rate is volatile, so the player’s own season carries only about 35 percent of the weight, and the baseline accounts for the other 65 percent. For instance, by considering 1.5 and 2.3 with weights 0.35 and 0.65, you will find a weighted average of 2.02, which is not far from the league average.
That is the whole mechanism. It is not a hedge or a way of dodging a call. It is the arithmetic consequence of the fact that some numbers repeat and others do not — the same logic behind the broader case for numbers driving roster and coaching decisions rather than gut feel. For a stable statistic like sack rate, the weights would flip, and most of the projection would come from the player himself.
Reading the arguments you are about to have
Two consequences follow. First, be careful with breakout seasons built on unstable statistics: a leap driven by a halved interception rate is a much weaker bet than one driven by a lower pressure-to-sack rate. Second, this is the right way to read any top-ten quarterback ranking. Moreover, as you can see, lists built on multi-year stable inputs age well, while lists built on one volatile season look strange within a year.
The 2025 season made the point. Sam Darnold entered the year on his fifth NFL team and by February sat one win from a distinction no starting quarterback had ever held. As the pre-Super Bowl LX rundown of records and milestones on the line in Santa Clara laid out, Drake Maye had a chance to become the youngest Super Bowl-winning starter in history in the same game. Which parts of that were signal is a question the next three seasons will answer — and the stability of the underlying statistics is the only honest way to guess in advance.
