Dynasty, in Practice: How Do Quarterbacks Age?
In an article last month, I made the case that we’re thinking about age in entirely the wrong way. The traditional thinking on aging is that it resulted in a steady and inevitable decline. After digging through 30 years of NFL history, however, I found that the “steady decline” model did not accurately represent what was actually happening to players on the field.
Instead, a better way to think of player aging is as a stable equilibrium prone to dramatic, sudden, and unexpected drops. Players essentially remain productive until one day when they’re not any more.
While I laid out evidence in favor of a new model of player aging, I didn’t actually provide specific data on what that model might look like. Today, I’ll examine 30 years of fantasy history to design a model that best represents observed aging patterns for quarterbacks.
Before I get to the results, I want to first discuss the process and its limitations. If you’ve already read my piece on receiver aging or running back aging, this will look familiar, so feel free to skip down to “Unique Challenges Facing Quarterback Models”. If all you care about is the numbers, you’ll find them in a table down at the bottom of the article.
A Quick Primer On Terms
Instead of an “aging curve” model, where I predict how much players will improve or decline at any given age, I will be creating a “mortality table” model. Mortality tables are used most prominently in life insurance, where insurers want to make educated guesses on how likely a person is to die over any particular time span.
It’s important to note that this model is not actually making predictions about the players involved. Instead, I’m merely measuring observed patterns among players who were superficially similar. If 18.6% of players who were superficially similar to Tony Romo “died” at age 33, that doesn’t mean Tony Romo has an 18.6% chance of “dying” this year.
This is really easy to demonstrate conceptually. Brett Favre is among the pool of players who are “superficially similar” to Tony Romo. If Brett Favre had torn his rotator cuff at 35 and never played again, the pool of comparable players would have a higher observed “death rate”. But would Brett Favre tearing his rotator cuff in 2004 make it more likely that Tony Romo declined today? Of course it would not.
Similarly, if Phil Simms hadn’t fractured his foot and lost his starting job, it wouldn’t make Tony Romo less likely to get injured or benched today. These are different players whose careers are wholly independent, and what happened in one case really has no bearing on what will happen in the others.
Likewise, if the model says Tony Romo has 2.26 more “Expected Years Remaining”, this doesn’t mean it thinks Tony Romo in particular has 2.26 years left. It’s not predicting that he’s going to fall off a cliff halfway through the 2016 season. Instead, it’s saying “if you had 1,000 quarterbacks like Tony Romo, then you’d expect the average remaining fantasy-relevant career length from the sample to be a bit over two years”.
Even that 2.26 average isn’t a specific prediction. We would expect only 39% of the 1,000 quarterbacks sample to “die” in either 2 or 3 years. We’d anticipate that 249 of those passers would still be productive at 38. Four of them would make it all the way to 41.
This is very important: we are not predicting specific outcomes. We are simply measuring risk. I don’t know how Tony Romo’s career is going to play out. He could be the next Warren Moon, making a Pro Bowl in his first year on a new team at 41. He could be the next Phil Simms and spend the rest of his career backing up younger players. He could fall off the planet, then re-emerge in his late-30s like Kurt Warner.
There are a lot of possible paths Tony Romo’s career could take. Nobody knows what will happen, and anyone saying otherwise is misleading you. I’m not trying to make a prediction, I’m simply trying to clearly outline the risks so that owners can make an informed decision. I’m not dealing in certainty, I’m trying to quantify the uncertainty.
A Note On Methodology
One problem anyone will have to grapple with when dealing with aging is called survivorship bias. In short, it states that the survivors of a process are not necessarily representative of all members of the process.
If I want to model how 42-year-old quarterbacks age based on historical comparables, there’s only one player I can compare to. Warren Moon is the only passer to start 10 games at 42 or older. The problem is that Warren Moon is an outlier, so how he aged is not going to be representative of how some other quarterback will age.
To a large extent, the advantage of a mortality table model is that it’s self-culling, allowing us to sidestep the survivorship bias. We’re only comparing players to their peers once they actually reach that age. Warren Moon is probably not going to be representative of how a typical back will age… but once a quarterback reaches age 42 in the first place, we already know he’s not a typical player. Suddenly Warren Moon becomes a much more reasonable point of comparison.
This culling effect really kicks by the mid-to-late 20s. By that point, any player who is still producing fantasy value is undoubtedly on his second contract in the NFL. He’s managed to stick around for half a decade. He’s probably a pretty good player.
In the early 20s, finding good points of comparison is substantially harder. If you looked at all 21-25 year old backs, the observed bust rate is going to be massive. Why? Because the majority of young passers aren’t any good. Included in that sample are a lot highly-drafted busts and late-round emergency fill-ins.
Obviously this doesn’t represent a good group of comps for someone like Andrew Luck or Russell Wilson. We already know that those passers are pretty good. I’d be willing to bet that, barring catastrophic injury, both players will still be a starter somewhere in the NFL after their 30th birthday.
In order to generate decent comps, I limited myself to looking at the top 50 retired fantasy quarterbacks since 1985. (The reason I’m only considering retired players should be self-evident.) This generates a pretty good list of guys who are largely “second-contract” type players.
This method is not without its flaws, though. Unlike running back or wide receiver, quarterback is not a position that lends itself often to players who look great early but then flame out quickly. Perhaps Robert Griffin III III will one day serve as one such example, but from 1985 to 2014, they were few and far between.
One thing that is a flaw, though, is that the conclusions from this method really only apply to guys who we already believe are “second-contract” type players. If I say that the average 25-year-old quarterback has 9.46 expected years remaining, I don’t mean that Mike Glennon, 25-year-old backup for the Tampa Bay Buccaneers, has 9.46 expected years remaining.
The general lack of interest in Glennon leaves us unsure of whether he’s any good, but inclined to bet against it, so a better set of comparables would be the larger list that includes all of those failed rookies who never received second contracts. Because right now, that career path is still very much a viable possibility for Glennon.
In short, be careful when using these EYR values. They’re really only meant to apply to players who we already have strong reason to suspect are probably pretty good. Even highly-drafted quarterbacks like Jameis Winston or Marcus Mariota are unlikely to be good candidates for this comparison. Later round players like Derek Carr or Jimmy Garoppolo would find their career prospects dramatically overstated by this methodology.
One final word of caution. These numbers model what happened between 1985 and 2014. They do not account for the possibility that things are dramatically different today. Improvements in modern medicine, for instance, could very well extend the careers of aging quarterbacks. I would not be surprised if these EYR numbers underestimated the remaining careers of 35+ year old passers to some degree or another.
Unique Challenges Facing Quarterback Models
The wide receiver and running back models were written up first because they’re generally the two cleanest positions for this sort of analysis. Large numbers of weekly starters and low positional baselines leave us with a very large sample size of former players to compare against. Because NFL teams play fewer quarterbacks per week, (typically just one compared to two or three running backs and up to five wide receivers), our sample pool of potential comparisons is smaller.
Additionally, since fantasy leagues start fewer quarterbacks per week, the positional baselines are higher, and quality quarterbacks are more likely to miss the cut in any given year for reasons that have nothing to do with a decline in play. As a result, while running backs and wide receivers tended to produce positive value over every year until their final fantasy-relevant season, quarterbacks pop in and out of relevance. This makes pinpointing a “decline” very difficult.
A great example of the difficulty of modeling quarterback decline is Kurt Warner. There’s really not an analogue to Warner at running back or wide receiver. Warner was a star until 30, irrelevant from 31 to 35, then a star again from 36 to 38. What do we mark down his “death” as? Or do we say he had two “deaths”?
Warner is hardly an anomaly among quarterbacks. Boomer Esiason, a former league MVP, was a fantasy starter from age 24 to 29, again at age 32, and then once again at age 36. Randall Cunningham dissappeared for three years before putting up a top-2 finish at age 35. Drew Bledsoe had a separate 3-year gap and 2-year gap between stints of fantasy relevance.
I mentioned that modern medicine might be extending careers, and if that’s the case, we’d expect the results to show up most strongly at quarterback. Careers are already longer than at any other position, and quarterbacks can do the most to get around an erosion of physical skills. We’ve already seen this from Peyton Manning, coming back from an injury that no player had ever before returned from and winning league MVP despite flagging arm strength.
Finally, there’s one last challenge inherent to mortality tables themselves. Mortality tables typically end eventually, rather than continuing on forever. The problem is that we never reach a certain age where a person has a 100% chance of dying. There are several methods for getting around this, the easiest of which, (and the one I have chosen), is to manually force the numbers to 100% at some point. At quarterback, I have set the tables so that the estimated chance of “surviving” past the age-42 season is 0%. This is a useful fiction that enables these tables in the first place.
A consequence is that any players who are near the arbitrary endpoint will have their EYR artificially reduced. At running back and wide receiver, no player is particularly close, so this is not much of a concern. The closest wide receivers are Reggie Wayne, who is likely to retire, and Steve Smith, who has announced his retirement after this season. The closest running back is Fred Jackson, who was just cut by Buffalo and is likely playing out the remainder of his career as a backup in Seattle.
At quarterback, though, Peyton Manning, Tom Brady, Drew Brees, and possibly even Tony Romo and Eli Manning are starting to get within shouting distance of our arbitrary endpoint. I would not be shocked if one of those five managed to make it to age 42 or even beyond. I wouldn’t bet on it, but to whatever extent it is possible, those players will be underrated in this analysis.
The sum of these biases leads me to believe that this approach is probably underrating older quarterbacks. I’m posting the data as calculated, but I felt it was important to lay out the challenges and let the user mentally adjust it as he or she sees fit.
To be clear, this is not to suggest that I do not believe that quarterbacks are a poor candidate for a “mortality table” approach to estimating future fantasy value. In fact, a piece on quarterback aging was what first convinced me of the merit of a mortality table approach. Quarterback rate stats show the exact sort of equilibrium punctuated by a sudden decline that we’re looking for.
The real underlying problem is that, from 1985 to 2014, there was only a weak correlation between quarterback quality and quarterback volume. The resulting data is messy, but I have the utmost of confidence in the underlying principals.
Enough Talk, Let’s See Some Numbers
Based on a best-fit curve covering all relevant passers over the last 30 years, this is how quality quarterbacks age in the NFL. DR% stands for “Death rate”, and measures the chance that a receiver at that age will suffer a catastrophic and career-ending decline.
EYR stands for “Expected Years Remaining”, and represents a weighted average of remaining career lengths based on observed data. The estimated career length refers only to how long a player is expected to remain fantasy relevant— players can and typically do play several additional years at the end of their career where they are still on an NFL roster but no longer producing startable fantasy value.