Last week, I looked at the question of when hitters hit their peak and came away with the very unsatisfying answer of “it depends.” Good hitters peak later. Even leaving aside the usual cries of “treat everyone as an individual!” there are different peak ages in the aggregate among different groups. The details matter. Now we are left with this question: What happens with pitchers?

First, let’s grab a sample. Like last time, I looked for all players who began their career after 1980 and ended it before 2009. I also looked only at starting pitchers, defined as a season in which 90 percent of the pitcher’s appearances were starts. The pitcher also had to face at least 250 batters in the season. This gave me a sample of 455 unique players.

The reason I went only with starters is that when a starter falters, there’s a decent chance that he’ll be given a shot at working out of the bullpen. It’s very rare that a pitcher makes the reverse conversion, and it’s usually only in the context of a dominant reliever whom the management feel would be better served by going into the rotation. It’s clear that teams consider the bullpen a drop down in rank. So, if a pitcher falls away from starting, we can consider that he has passed his peak. I know that won’t sit well with everyone, and I’m sure we’ll have a lovely discussion down below. But for the purposes of this study, once a pitcher is no longer a starter, he no longer counts. (Yes, yes, he might make a dandy reliever. It worked for Dennis Eckersley.)

One thing to know is that starting pitchers tend to debut (that is, have a season where they faced 250 batters or more) later than hitters. Hitters had an average age of debut of 24.63, while pitchers tended to take their bow at age 25.05. Like hitters, pitchers generally debuted between ages 23-27, with a smattering of players on either side. Both most commonly debuted at age 24.

Pitchers, it seems, are given a little more time to perfect their skills before being called up to the majors. And it makes sense. There’s a certain urgency about starting pitching in the game. Games are often subtitled by their starting pitching matchup (It’s Tim Linececum vs. Roy Oswalt tonight!) There’s the old canard about “you can never have enough pitching.” And it makes sense. If a hitter has a bad night, he’s just one out of nine on the team. If a starter has a bad night, it just about guarantees a loss. So, teams (and fans) are a little more skittish about their starters.

But when do pitchers tend to flame out (or at least, stop being used as starters?) The most common age of exit in this sample was 27, much like the hitters. There was a small secondary peak at age 30 (the hitters had a much more well-defined peak at 31), but this peak for the pitcher wasn’t very well defined. When I looked at the age of exit, broken down by age of debut, a very simple pattern emerged. Pitchers were overwhelmingly exiting the talent pool after their first year. Thirty-eight percent of all starting pitchers who had a season in which they faced 250 batters did not have another such season.

While with hitters, the age-27 and age-31 “hurdles” appeared in almost all age-of-debut groups, there didn’t seem to be any such coherent trends for when pitchers left the game (or, at least, the starting rotation) other than their first year. Since pitchers who only pitch one year, by definition, peak in that year (at the MLB level anyway), I threw them out.

For the rest of the sample, I tried much the same methodology with the pitcher sample that I did with the hitters to determine peak age. (Quick recap: Mixed linear model, age as a fixed effect, AR(1) covariance matrix… it’s all in the last article.) I used FIP as my outcome variable. I was hoping for a chance to use the newly-developed SIERA, but that requires batted ball data, which wasn’t available as far back as I needed it. Like last week, I looked for peaks in each debut-age group.

The results were… less than clear. In most groups, there were multiple peaks, but there didn’t seem to be much of a pattern. Pitchers who debuted later, assuming that they stuck around for more than one year, tended to peak later on their group’s second peak, but even those didn’t really line up nicely.

Debut Age  Peak(s)
   0-22    26, 29
   23      26
   24      29
   25      25, 29
   26      27, 32
   27      32

I also looked at whether age of debut was associated with a) flaming out of the starting rotation after one year or b) being invited to stick around in the bullpen for a while (defined as a relief “career” of at least three years). A player can do both (he starts for one year, then moves to the pen for three years.)

Debut Age  One-year Wonder  Became Reliever  
   0-22         22.4%           40.8%
   23           34.9%           36.0%
   24           38.3%           51.9%
   25           33.8%           36.6%
   26           47.1%           43.1%
   27           51.4%           40.0%

In general, we see the same sort of pattern for age of debut and one-year flameouts that we do for hitters. The younger you are when brought up, the more likely you are to stick around for more than a year. There’s a pretty sharp dividing line between age 25 and age 26. Why the dividing line between 25 and 26? There’s no inherent reason for it to be that big, but it suggests that teams view 25 as the expiration date for pitching prospects. If you’re any good, the team will probably bring you up by age 25. If you’re a starter, 26, and still in Triple-A, you may want to start thinking about another career.

As to who become relievers, cover up that line for age 24 for a moment (a fluke?) and we see that about 40 percent of failed starters from all age groups try their hand (perhaps their left hand?) at relieving with some modicum of success. Perhaps there’s more of a profile-guys with two (and only two) good pitches?-that teams look for when shifting guys to the bullpen, and that isn’t associated with age of debut. But what’s up with age-24 debuts going to the pen? Perhaps some of these are guys who didn’t quite pan out as phenoms, but who the team just doesn’t want to give up on. Oddly enough, their brethren who debut at age 25 fare better in not flaming out after one year.

So when do pitchers peak overall? Last week, I came to the conclusion that while there wasn’t one definable “peak age” for hitters, there were at least some guidelines that could push us forward in the investigation. (Translation: at least there was generally just one peak age per group.) Here, I have to admit that I’m at a loss. With multiple peaks and valleys, pitchers are a completely different animal than hitters. Perhaps different types of pitchers (power pitchers, junkballers, middle-of-the-road guys) peak at different times. Maybe different body types age differently. But while hitters seem to mature in somewhat predictable ways, pitchers seem to require a complete root canal in terms of our understanding of how they age.

Thank you for reading

This is a free article. If you enjoyed it, consider subscribing to Baseball Prospectus. Subscriptions support ongoing public baseball research and analysis in an increasingly proprietary environment.

Subscribe now
You need to be logged in to comment. Login or Subscribe
It seems there is somewhat of a trend in which guys peak a certain amount of time after they make their major league debut. This seems consistent with the hitters as well and is completely logical. Perhaps it would be useful to categorize players first, not by role but by stuff -- perhaps handedness and average fastball velocity (the latter as more of a proxy for the complex variations of pitch types and velocities).
I'm beginning to wonder whether modeling age directly is the most informative approach to understanding baseball careers. Specifically, why is AGE a dependent variable in your analysis rather than an INDPENDENT one?

Suppose you set out to model the major league career (ML) as your DEPVAR (CareerMonths--CarMos), and then used Age as a factor in predicting CarMos, along with other predictors including performance indicators, mos. lost to injury, (for pitchers) total batters faced in career, position, teams played for, and some global factors (era played in, league played in, DH rule in place, etc.). It could also take certain "phenotypical" characteristics into account, including handedness, BMI, speed (analogous to the PECOTA model).

I think this approach might be more consistent with how teams make decisions to promote/demote, renew contracts, and so forth. They may take age into consideration as a factor in their decisions, but age is just one of many factors that would thus contribute to the expected CarMos of a player, controlling for performance, injuries, and other Indepvars.

This approach would also lend itself to applying an explicit survival model to careers. And it could even be extended to a multistate model, in which the states of a career begin from the first year a player is drafted or signed, and then changes "states" from unemployed to employed or back again, from minor league to major league or back again, to retired or died. Such multistate modeling is commonly used for certain kinds of changes in work status or health status, for example.
Even you don't use a survival model as such, or aren't interested in a multistate model, you might use some form of regression model specified along these lines -- with the aim of isolating the effect of aging on career length, as well as age at first entry to the major leagues, controlling for performance and other factors:

(1) You could model as your depvar the total number of career person months in the major leagues (TOTMLMOS) as a function of age at first callup, performance, position, phenotypical factors (body type, speed, handedness a la PECOTA), injury, interruptions (due to injury or demotions or war service, etc.), and performance (e.g., using Warp, or perhaps offensive value and defensive value variables separately).

(2) Or you can model expected future major league months (FUTMLMOS), as a function of prior cumulative MLmos, age, performance (cumulative lifetime Warp, as above), injury, and interruptions.

The functional forms of the predictors can take various forms, e.g., a nonlinear form for the age variable or you can test for whether there are critical breaks in the effect of age on future MLMOS, controlling for cumulative performance or the previous year’s performance or previous 3 year baseline.
I'm not a huge fan of articles that can't draw to a conclusion. While that's not your fault, maybe that's an indication that with this particular topic, you're chasing the Questing Beast.
I'm not a fan of writing these types of articles either. I like to have a punchline at the end too. If there is a conclusion to be drawn, it's that just because a methodology works in one situation, it might not work in another.
Pizza good work -- I your numbers seem to fall in line with some work I have done. I was more looking for the year that an established starting pitcher is done and I found age 32.