In the midst of an offseason best characterized as “monotonous,” Jose Altuve’s extension stands out. A week ago, Altuve signed a 5 year $125 million deal with the Astros, cementing his future with the franchise that signed him 17 years ago. While Altuve (and the Astros) may never excise the taint of the 2017 sign-stealing scandal, his diminutive stature, unlikely success story and exemplary career combine to make him one of the more memorable players in baseball.
The Altuve deal, as Michael Baumann of Fangraphs points out, may be more about sentimentality than pure player value alone. Altuve’s extension begins in 2025 and covers his age 35–39 seasons. The ZiPS projections for those seasons are not optimistic, totaling just 6.3 WAR, which would not be close to requisite production for a $125m salary. It’s fair to believe that Altuve may best that pessimistic outlook, given his unique profile and recent performance. However, while researching the deal, I noticed something interesting. 2023 was not exactly a banner year for older position players. In fact, no position player in their age 36+ season recorded 2 or more fWAR…
To put this fact into context, it is the first time (including shortened seasons) that this has happened since 1963, when Forrest “Smoky” Burgess maxed-out at 1.8 fWAR. Furthermore, just 3 players age 36 or older recorded at least 1 fWAR in 2023 and only 14 received >100 plate appearances.
These facts may be interesting but they are hardly dispositive. 36 years old is a somewhat arbitrary threshold and 2023 is potentially an anomalous season. Taking a broader view though, there does appear to be a trend of decreased success for older position players. Here is how 2+ WAR seasons have been distributed by age since 2005 (the “post-steroid” era)…
Notice the relative decrease in 2+ WAR seasons from 35 and older position players compared with other age groups in more recent seasons. This is true in terms of the share of total WAR accrued by the oldest position players as well…
A decrease in WAR value can generally be attributable to two things: 1) A lack of playing time 2) Worse production on a rate basis. In this case, the 35+ position player cohort has suffered from both of these issues of late…
Clearly, older position players are just not performing at the same level in recent seasons as they did 1–2 decades ago. The drop-off in both performance and playing time is particularly jarring. Typically, it would be expected for there to be an counterweighting equilibrium between those two facets. Struggling veterans can drop out of the player pool by retiring, leaving a smaller base of better-performing older players. Yet rate performance (represented as WAR/500 PAs) has also decreased for 35+ year olds. Veteran bats are simply not performing near the level they used to.
Aging Curve Methodology
Of course, analysis of the performance of older players is incomplete without a look at the aging curve. Jeff Zimmerman, Mitchel Lichtman and others have written extensively on the impact of aging on player performance, both generally and for specific skills and cohorts of players. The most popular method for determining an aging curve is the delta method, which compares individual player performance (with a playing time minimum) in back to back seasons (and ages), weighted by playing time. The harmonic mean of these weighted differences in performance is then used to construct the aging curve. This process is described in more detail here. In 2020, Zimmerman used the delta method and found that the hitter aging curve had indeed steepened for several key statistics between 2005–2011 and 2012–2019.
While the delta method works quite well, in this case I’ll instead be using a regression model, specifically a Generalized Additive Model (GAM), with piecewise smoothing for age. Regression models have also been used to model aging curves in the past, including in work by JC Bradbury, CJ Turturo (in hockey) and most recently Jonathan Judge. In non-technical terms, the GAM works like a flexible ruler, smoothing smaller (piecewise) lines into a curve designed to fit trends present in the data, in this case performance by age. However, as Phil Birnbaum notes, using a regression of purely performance vs age to determine the aging curve is dangerous. For instance, a GAM of purely age vs wRC+ yields this rather flat specimen…
The issue here is one of selection bias, as the player pools at each age are not identical. Players that remain active at older ages tend to be better as a whole (over the course of their careers) than those active in their late 20s. The weaker-performing players simply decline themselves out of a job as they age. This selection bias masks the aging process. There are several ways to address this issue (such as using random effects regression or factor smoothing for individual players) but a simple method used by Judge is to include career mean performance (for each player) as a fixed effect along with age. This should significantly negate the selection bias issue by setting a “baseline” to separate overall player quality from the effects of aging. Indeed, Judge found that a GAM of this kind (with each player-season’s influence weighted by plate appearances) matched or bettered the performance of the delta method in multiple tests. Regression can also handle players who miss seasons more organically and is a bit easier to compute. For those more technically inclined, the GAMs used (in line with Judge’s method) follow this template:
model <- gam(stat ~ s(Age) + career_stat, data = dataset, weights = PA)Aging Curve Analysis
Aging can affect multiple facets of a player’s game, but top of mind is of course overall batting performance. Here is how the wRC+ aging curve stacks up for the season ranges of 2005 to 2014 (largely post-Bonds/steroids) versus 2015 to 2023…
Similar to Jeff Zimmerman’s findings in 2020, wRC+ has seemed to decline at 30+ years old more rapidly in recent seasons than 1–2 decades ago. Also of note (and in contrast to Zimmerman’s curve), is that young players in the past ~10 years appear to be entering baseball closer to their offensive peak, rather than steeply ascending in their early 20s.
Changes in aging tendencies can also be reflected in underlying batting components. The aging curve for K% has changed significantly for older players, while the corresponding curve for walk rate has not…
This also fits with Zimmerman’s 2020 study, in which he posited that the increase in the steepness of the K% aging curve was due to an increase in fastball velocity and decrease in breaking ball usage by pitchers. More succinctly, veteran batters have had trouble making contact of late…
What about results on contact? Here are the aging curves for BABIP and ISO in each time period…
It appears that the aging curve for ISO may be a bit flatter in recent seasons, while BABIP may be aging more rapidly, though the differences are subtle and are largely within standard error at both ends of the curve (where the sample is smaller).
Last but not least is defensive aging. In this case I’ll be using Fangraphs’ “Def” metric on a per 600 plate appearance basis. Def combines both a measure of fielding value and a positional adjustment. Here are the Def per 600 PA aging curves for both time periods…
It appears that defensive value has declined far more rapidly after the age of 35 in recent seasons, by a measure of ~1–4 runs. Anecdotally, this seems to fit with the preponderance of designated hitters and/or first baseman amongst veterans in 2023. Several prominent elder statesmen, including Andrew McCutchen and Justin Turner, have largely transitioned to DH in the past few seasons. Generally, without exemplary hitting performance it is difficult to be an impactful contributor while providing meager defensive value.
Key Takeaways
- The overall value (measured as fWAR) produced by veteran position players 35 years and older has decreased in the past 5–10 years.
- This is both due to a decrease in playing time and production on a rate basis, as measured by fWAR per 500 PAs.
- The aging curve for batting production seems to have changed in the past ~10 seasons, as younger players are peaking a bit earlier and veterans are declining at a faster pace.
- This shift seems largely attributable to changes in the aging curves for K% and Contact%, with the possibility of smaller and less certain differences in aging for results on contact.
- The aging curve for defensive value seems to have steepened for veteran players in the past 5–10 years, particularly for those 37 and older.
Conclusion
Trends in MLB are constantly changing. Sinkers were all the rage, declared dead and now may be back, all in the span of a decade. Developing and acquiring young hitting talent is hardly a new phenomenon, but it is quickly becoming a necessity. The performance of older hitters may rebound, but at the moment they do not seem to be the wisest investment. Returning full circle, Jose Altuve’s extension was about more than just value and on-field production. But if Altuve can continue his winning play, he’ll be defying expectations once again.
Thank you for reading this post. Much of the R code used to generate both the plots and aging curves can be found here. Thank you to Jeff Zimmerman, Jonathan Judge, CJ Turturo and many others whose previous work on aging curves was extremely helpful. I’d appreciate any positive or negative feedback and can be reached both on Twitter and at m26762059@gmail.com.
