# Pitcher Aging Curves: Introduction

As on-field performance data has evolved, baseball enthusiasts have been spoiled with more precise measures of player performance. One area in particular is pitcher velocity. Whether through Baseball Info Solutions (BIS) or PITCHf/x, writers and researchers can now add a critical variable into their analysis that wasn’t readily available a decade ago.

Many readers of FanGraphs and Beyond the Box Score have seen Jeff Zimmerman‘s position player aging curves. After reviewing them, I started to pester Jeff to see if he considered similar curves for pitchers â€” specificallyÂ in the area of fastball velocity. I was curious about the general pattern of decline for fastball speed and how it impacts overall pitcher performance.Â Luckily, Jeff already had been thinking about this.

Today, Jeff and I are launching a multi-part series on pitcher aging curves, which is centered on fastball velocity. This introductory article will lay out the methodology we used and â€” of course â€” the initialÂ baselineÂ curves for all pitchers, as well as starters versus relievers.

Here are the overall pitcher aging curves:

Let’s take a look at the methodology used to produce the curves:

For the aging curves, we compared pitcher seasons from 2002 to 2011 using fastball velocity (FBv) data from BIS. This means that we have more than just four-seam fastballs mixed into the sample for each pitcher. Ideally, we would have focused just on four-seam fastballs â€” but given that PITCHf/x only has been around since 2007 â€” this was necessary to get the volume of data required for the curves.

The same method was used as the one that Jeff previously used to calculate hitter aging curves (all the heavy math is in the linked article). To get the aging amount, Jeff took each of the pitcher’s rates in one year and compared them to how they did the next year. Each pitcher’s change was then weighted by the harmonic mean of the number of innings they pitched between the two seasons. Finally, all the weighted values are added together to get the total amount of change for each metric. Note that the curves are cumulative: For example, when it comes to fastball velocity, pitchers lose a total of 3.75 mph on average during their careers.

Here are the metrics we looked at for this study:

Metric Description
Velocity Fastball (FBv) speed in mph
K/9 Strikeouts per 9 innings
BB/9 Walks per 9 innings
LD% Line drives as a percentage of all batted balls
GB% Ground balls as a percentage of all batted balls
FB% Fly balls as a percentage of all batted balls
HR/9 Home runs allowed per 9 innings
BABIP Batting average on balls in play
SWG_Strike Swinging strike percentage
FIP Fielding Independent Pitching

From time to time, you’ll notice that some of the curves areÂ labeledÂ with “x 10.” This means that â€” for the purposes of visualizing these metrics side-by-side â€” we multiplied theÂ cumulativeÂ change by 10. Had we not, those curves would have basically looked like straight, horizontal lines.

Some initial thoughts:

â€” Velocity is a young man’s game. Rather than a parabolic curve of some sort, pitchers generally lose velocity from the beginning. Through age 28, they appear to stay within .5 mph of their peak velocity; but starting at age 29 they have lost about 1 mph with the loss accelerating every year thereafter.

â€” The loss of velocity is important because we see that pitchers’ abilities to record strikeouts follow a curve similar to the speed of their fastballs. However, the slope of the decline is not as dramatic as the velocity decline. This is perhaps do to a couple factors. First, pitchers are likely to further develop secondary and tertiary pitches as they mature. Many of the best arms in the minors can dominate using mostly their plus-fastball and little in the way of plus- or above-average off-speed stuff. But surviving in the majors requires more than just a plus-fastball â€” which many pitchers quickly realize. Either they develop additional weapons, or they move on. Second, and somewhat related, pitchers might develop a fastball with additional movement â€” like a sinker or a cutter â€” to compensate for the velocity decline. This also could lead to a less steep decline in K/9 and SWG_Strike rate than a pitcher’s velocity decline alone might predict.

And what about batted ball type? Behold:Â

Even with each batted-ball-type percentage multiplied by 10, it’s a bit tough to see. Essentially, what we observe is that as velocity decreases batting average on balls in play (BABIP) increases. This is because as velocity goes down, the percentage of line drives and fly balls increase at the expense of ground balls. Our guess is that while ground balls generally have a higher BABIP than fly balls, the fly balls being hit off lesser-velocity pitches are likely better hit.

We also observe an uptick in line drives during a pitcher’s age-34 season. This uptick seems to align well with a similar jump in BABIP during a pitcher’s later years.

These initial curves incorporate both starters and relievers. Obviously, this could (and does) introduce some bias into the data, since ineffective starters are unlikely to remain starters as they age. This could artificially lessen the slope of the velocity curve. Don’t worry, though, we have the data broken out by starters and relievers and will be addressing those issues in later articles.

For now, here are the overall aging curves for starters and for relievers:

Starters were coded as pitchers who threw more than 80% of their innings as starters in both year one and year two. For relievers, we used the cut-off of greater than or equal to 66% of innings in relief for both years.

The curves are obviously quite different for both classifications.

First, change in reliever velocity is more tightly aligned with change in strikeouts. Starters seem to be able to better maintain their strikeout rates even as their velocity declines; relievers appear more dependent on their fastball speed to be effective. Here’s that difference in chart form:

Second, walks trend up from the get-go for relievers. So, too, do home runs and batting average on balls in play. In fact, reliever performance in general worsens at a much faster rate than starters, except for velocity. Looking simply at the Fielding Independent Pitching (FIP) curves, between ages 23 and 29, starters on average experience a .54 increase, while relievers average a 1.20 increase.

Let’s take a look at two starting to pitchers to help illustrate the curves. Roy Oswalt and Barry Zito were both 24 years old in 2002, so we have a almost a full career’s worth of data to plot. When we compare their drop in velocity and strikeout rate against our average for all starters, we see that both players follow the pattern we established earlier:

While Oswalt and Zito aged at different rates, the relationship between their velocity and strikeout rates is similar to our average across all starters. During a starter’s age-33 season, the strikeout rate will decline roughly .55 for every 1 mph the pitcher lost in velocity. Oswalt lost about .4 strikeouts per nine innings for every 1 mph his velocity declined by 33; Zito’s strikeout rate fell by .59 for every 1 mph he lost.

During the next few weeks, Jeff and I will publish more articles based on the data and our findings. For example, we’ll look at starters and relievers in more detail â€” as well as the differences between the two. We’ll also look at what happens when pitchers maintain their velocity from year to year, and how other pitchers deal with velocity losses.

There are lots of things to discuss and untangle, so we would love feedback on the research that’s already done, as well as thoughts about additional analysis.

Print This Post

Bill works as a consultant by day. In his free time, he writes for The Hardball Times, speaks about baseball research and analytics, consults for a Major League Baseball team, and has appeared on MLB Network's Clubhouse Confidential as well as several MLB-produced documentaries. Along with Jeff Zimmerman, he won the 2013 SABR Analytics Research Award for Contemporary Analysis. Follow him on Tumblr or Twitter @BillPetti.

Member
Richie
4 years 27 days ago

Has anyone ever looked at curve ball velocities and ever found anything suggestive of anything? (slower curve balls break more and hence are more effective; faster curveballs break sharper and hence are more effective; curve ball velocity or perhaps use fluctuates over time and correlates in such-and-such a way with effectiveness; and so on)

Guest
Dan
4 years 27 days ago

I agree that research would be very interesting.

Member
payroll
4 years 27 days ago

I did a very rudimentary calculation of sliders and cureball weight values, as compaired to average velocity. There is a positive relationship of weighted value to velocity with breaking pitches, but it doesn’t appear to be as closely related as fastballs are.

Guest
Andrew
4 years 27 days ago

My understanding of these curves remains that they’re underestimating the impacts of age. If I have two 34-year old pitchers, but next year one retires and the other continues to pitch, it’s pretty likely the one that retired ‘decayed’ more than the one who still throws, e.g he no longer has a job because his fastball slipped to 86 instead of 88 over the offseason. But the guy who now throws 86 isn’t observed, and gets dropped from the sample.

Guest
Spencer
4 years 27 days ago

Totally agree. There’s probably a lot of sample mortality issues (players retiring) in the later years.

Member
jfree
4 years 27 days ago

There’s an element of the reverse issue though as well. The young fire-throwers are more likely to enter the league earlier than the tepid-throwers (who may have to master a second pitch or spend time improving control/command in order to get their ticket punched to the show). Even looking at the anecdotal – Oswalt clearly showed no velocity decline until his early-30’s; Zito’s velocity simply looks more erratic (with a clear decline correlated with the size of his contract)

Guest
Paul
4 years 27 days ago

The same inherent problem exists in Zimmerman’s hitters series. Tango, et al. used a different methodology than Bradbury for that analysis, which I disagree with. But they at least confronted the inherent issues in their model by explaining the specific methodology they used in the analysis to try and correct for that deficiency.

So you just have to know going into this that if Zimmerman’s involved that inherent problem is going to be there. My problem with this approach is that for folks like me, and apparently some others, while I like the idea and I think the analysis would be interesting, it’s not even wrong and merely trivia that I won’t much bother with. This sounds like an attack on Zimmerman, but it’s really not. It’s just that when you have fundamental differences, you’ll have separation and disunion.

Knowing that that was going to be an issue, my bigger beef is that FIP was used here, instead of either ERA (just as reliable over time), or SIERA (even more reliable and most powerful using smaller samples). Since SIERA is probably the most predictive metric we have (hitters or pitchers), it’s a shame that it was not used for an analysis that it is perfectly suited for, and since it has good predictive power for smaller samples, might help correct for some of the fundamental problem noted in the original comment.

Guest
Blue
4 years 27 days ago

Nice. The difference between relievers and starters is pretty straightforward I suspect–your most talented pitchers will, by and large, become starters. Even a small degradation in talent level among your relievers can lead to them falling off a cliff.

Member
Redscot
4 years 27 days ago

True Blue. Not to mention often times starters get converted to relievers because they have not developed a 3rd pitch, which would further exacerbate their problems with a declining fastball.

Guest
MikeM
4 years 27 days ago

Very cool. Thanks guys.

Member
4 years 27 days ago

Nice job. I don’t mean to be petty, but these graphs are hard to read. The colors are not very distinguishable, all the lines are the same width and shading (except at the end) and I don’t know why they’ve been printed on a grayish background–that makes it even harder to distinguish the lines from each other.

Member
chuckb
4 years 27 days ago

You can click them and enlarge them and they’re quite easy to read. I think I agree that they’d look a little clearer against a white background but the larger graphs are very readable anyway.

Member
4 years 27 days ago

Go ahead and be Petti. This piece’s author certainly is.

Guest
Oliver
4 years 27 days ago

How curious that swinging strike rates seem to tick up modestly, but K/9 rates implode.

Guest
kds
4 years 27 days ago

Don’t sinkers and cutters average a bit slower than four-seamers for the same pitcher? Maybe older pitchers are choosing to throw a higher % of cutters/sinkers, making it look as if they have lost more velocity than is actually the case.

Andrew correctly raises issues of bias in the data. Might some of the observed differences between starters and relievers be due to differences in quality and choices of who starts and who relieves. And might these work differently at different ages.

Member
4 years 27 days ago

You’re right. Sinkers and cutters will generally have lower velocities, but there’s currently no public source of trustworthy (or at least consistent) pitch classifications going back multiple years, to separate that out.

This introduces all sorts of other problems, but I’d like to see these aging curves looking at the top some percent of fastballs in fastball velocity. That might be a quick way of getting mostly four seam fastballs and comparing apples to apples.

Guest
Hurtlockertwo
4 years 27 days ago

I think it’s obvious (especially to us who have lived it) that you are not as athletic as you get older. I would be interested in seeing the data to show that even with decreased velocity a pitcher can still be effective. After all, older pitchers still face a steady stream of young guys even as they age.

Guest
Jamie Moyer
4 years 27 days ago

Is there any reason you stopped these curves at age 37? Should go out a lot futher.

Guest
Jake
4 years 27 days ago

I think one thing that should be addressed in the comparison between a pitcher’s declining velocity and his declining K/9 rate is his increase in understanding of how to approach hitters. A nice comparison piece would be looking at guys who threw two, three, or four pitches throughout their careers. How did their velocity decrease? How did their K/9 rate decrease? My hypothesis would be that pitchers who threw more pitches saw a greater decline in their velocity, but a slower decline in their K/9 rate. It’s easier to set up hitters if you have more pitches, and that fact leads to greater room for improvement.

Guest
RC
4 years 27 days ago

I would love to see ERA on here instead of FIP.

FIP is only more predictive than ERA in pitcher’s parks, whereas ERA is more predictive in hitter’s parks.

EIther use SIERRA, or use both ERA and FIP. There’s no reason to have FIP on there when you already have K/9 and BB/9. Its just a conglomeration of the two.

Guest
RC
4 years 27 days ago

I’m also not a huge fan of adding more data to increase the sample size, when you know that data is less accurate.

Guest
Paul
4 years 27 days ago

2 for 2.

Guest
RC
4 years 27 days ago

If you’ve got a problem with my posts, and are intelligent enough to, please point out specifically what is wrong, instead of posting useless 3 word comments and falsly believing yourself clever.

Guest
Mcneildon
4 years 26 days ago

I think Paul was indicating agreement with you there. He had earlier voiced his disagreement with using FIP for this analysis. It appears that also agrees with your objection to adding data which may be unreliable.

Guest
Paul
4 years 26 days ago

Yikes RC, I was just trying to save some time there instead of putting a +1 under each of your comments that I thought were spot on.

Member
4 years 26 days ago

Paranoia paranoia everybody’s comin’ to get me

Guest
RC
4 years 26 days ago

I’m just used to getting jumped all over every time I mention that FIP is a pretty terrible stat.

Member
chuckb
4 years 27 days ago

This is really awesome! Thanks. I look forward to the next rendition.

Member
4 years 27 days ago

I think the sharper decline in skill for relievers, especially in the younger years, might be partly attributed to a bias in the data. With most positions, including starting pitching, young players usually need to impress in the upper minors before they’ll be given a role with significant playing time at the majors. For relief pitchers, it seems to me that teams use the throwing-spaghetti-at-the-wall approach a little more frequently. There are a large number of relievers who are thrown into modern bullpens as question marks, and either they prove themselves and stick or they’re replaced by more spaghetti.

Consequently, if a reliever has 2 years with a decent number of innings pitched, they probably pitched well enough in the first of those years to earn a secure role for the second year. If they pitched poorly, they probably were dropped, and while they might return to the majors, it may not be for a couple years. This would result in a larger measured decline in year-to-year performance than the underlying change in true talent.

This selection bias is common to all aging curves, but my hypothesis is that it is more pronounced in relief pitchers due to teams’ approaches in roster construction. I’d like to see this at least addressed in the relief pitcher article, though I’m not sure whether there is a clear solution.

Guest
4 years 27 days ago

Very interesting research. One thing that stands out immediately is that velocity tends to go into rapid decline at a pretty young age. So, if a pitcher is going to have any longevity, he needs to either start out with a very high velocity or quickly learn to pitch without velocity. I’d like to see how different kinds of pitchers age – ground ball pitchers, power pitchers, lefties, etc. One thing in particular I’d like to see is the difference in slopes between power pitchers and finesse pitchers.

Guest
Pat G
4 years 27 days ago

Ive always thought doing curves for pitchers is extremely difficult due to the ability for them to constantly be changing their repertoire/grip/sequencing… looking at FB velo is fine, but when you start looking at K/9, BABIP etc you are opening a whole can of worms. It’s a completely different beast from aging curves for hitters where their skillset doesnt change often.

Guest
Paul
4 years 26 days ago

Yep, key point. So what did we really end up with here? That pitchers do not throw as hard and are not as “talented” as they age. Maybe I’m in the minority here, but neither of those things surprises me in the least.

Guest
Mr Punch
4 years 26 days ago

It’s not just pitchers’ skillsets that change – it’s their roles as well. Starter, middle relief, LOOGY, closer are different jobs. Aceves has “gained velocity” as a closer, for example. I wonder if it might clarify matters to treat pitchers’ times as starter or closer as their “real” careers, when they have actual value.

Guest
4 years 26 days ago

This contribution is excellent and much appreciated. Look forward to the rest of your analysis

Guest
Dave
4 years 26 days ago

Is there information on how much velocity declines for a pitchers based on how much velocity they started with? I feel like (anecdotely) pitchers who start their careers off throwing high 90s see there velocities drop more than pitchers who start off in the low 90s.

Guest
Noah
4 years 17 days ago

“The loss of velocity is important because we see that pitchersâ€™ abilities to record strikeouts follow a curve similar to the speed of their fastballs. However, the slope of the decline is not as dramatic as the velocity decline. This is perhaps do to a couple factors. First, pitchers are likely to further develop secondary and tertiary pitches as they mature…Second, and somewhat related, pitchers might develop a fastball with additional movement â€” like a sinker or a cutter â€” to compensate for the velocity decline. This also could lead to a less steep decline in K/9 and SWG_Strike rate than a pitcherâ€™s velocity decline alone might predict.”

Wouldn’t a more important factor be the scale of the x-axis having inconsistant units? The slope for fastball velocity decline was in mph, but the others are showing decline in K/9, FIP, etc. All of them have different units so we shouldn’t expect them to have the same absolute value decline. Unless I missed something…

Guest
Keith Christoffers
3 years 3 months ago

Solid!

Guest
Neric
2 years 9 months ago

My take on these graphs is that they contain some artefacts from the steroid-era. Certain pitchers lose more than just their age-related velocity, they also lose the little extra once they go clean. Add it up and it should explain the huge drop that some guys experienced over the last 2-3 years.

Guest
tex
1 year 10 months ago

I have a question. Here, why use the harmonic mean vs. some other mean? Why does the harmonic mean make the most sense here? Thanks!