Component Changes in New Hitter Aging Curves
A few weeks ago I noticed hitters no longer improved, but instead their production only declined as they aged. Additionally, I answered at few questions on the first article in a second article. Today, I am going to look at the individual hitter traits to see which ones may be leading to the early decline.
First off, I will start off with some disclaimers.
1. For most of the two previous articles, I used wRC+ as the metric to determining aging. It is adjusted to the run scoring environments for stadiums and leagues. The following values in this article use raw datawhich is not league adjusted. As I stated in the first article:
A problem exists when using wOBA in the recent lower scoring environment. The league wOBA in 2006 was .337, and in 2013 it was at .318. That’s a drop of 19 points in seven seasons, or 2.7 points per season. Players will have the appearance of aging from season to season … This overall decline leads to a large year-to-year aging factor
So when looking at the upcoming data/graphs, remember the overall run environment has changed. Additionally, I have included the start and stop league values for the time intervals to help put the data change into perspective.
2. Note: The aging curves were created by the delta method which weighs the plate appearances using their harmonic mean. With this method, there’s a small survivor bias summarized by Mitchel Lichtman at the Hardball Times:
… survivor bias, an inherent defect in the delta method, which is that the pool of players who see the light of day at the end of a season (and live to play another day the following year) tend to have gotten lucky in Year 1 and will see a “false” drop in Year 2 even if their true talent were to remain the same. This survivor bias will tend to push down the overall peak age and magnify the decrease in performance (or mitigate the increase) at all age intervals.
OK, now it is time to look at some of the data. I decided to concentrate on four major hitter traits: walks, strikeouts, isolated power and BABIP
Walk Rate
1998: 8.7%
2005: 8.2%
2006: 8.4%
2013: 7.9%
Walk rates have been fairly constant over the 14 year time frame, so the curve can be compared without too much adjustment. The two curves are somewhat similar with the current seven year time frame peaking at bitter earlier at age 26 to 28. In the past the peak was near 29 to 30. This earlier peak would move the aging curve some, but not much.
Strikeout Rate
1998: 16.9%
2005: 16.4%
2006: 16.8%
2013: 19.9%
Strikeouts have exploded over the past few seasons with an over a 3% point overall increase. Looking at the curves, the early decline in strikeout rates to age 25 still exists. Previously, the strikeout rate stayed the same until a slow decline after a player’s age 30 season. Current players now see a 2% point increase in their K% from age 24 to 30. Just a handful of years ago no change would be expected.
The earlier increase would push the aging curve down more in the late 20’s, but again not enough to explain the early decline.
ISO
1998: .154
2005: .154
2006: .163
2013: .143
Finally we are getting somewhere. Players are now seeing their power decline almost immediately. Players no longer gain power, naturally or unnaturally, over their first few major league seasons. For example, home runs went from a peak of 5692 in 2000 to 4661 in 2013 (18% point drop). Besides the obvious PED elephant standing in the corner, I think players are more and more in shape when they reach the majors and they don’t have any more strength to gain.
BABIP
1998: .300
2005: .295
2006: .301
2013: .297
OK, now we are cooking with propane. Players now see their BABIP drop immediately. It wasn’t until age 26 when the decline previously occurred. I base the early decline on the increase knowledge and use of defensive information. Teams are doing a better job of tracking players in the majors and minors and know their hit tendencies. Teams are also more willing to make adjustment, such as over shifting a pull happy player compared to the past.
In the 2014 Hardball Times Baseball Annual, I examined extreme shift tendencies. Matt Adams hit into a shift 22 times in the first half of the 2013 season and posted a .368 BABIP. This value is unique because Adams isn’t really a feared hitter and only had 91 PA in his first season in 2012. Teams still decided he pulled the ball enough to implement a shift against him in the first half of 2013. In the second half, he hit into 39 shifts and his BABIP dropped to .308. Teams seem more and more will take a small amount of hitter information and position their defense accordingly.
The immediate decline in BABIP is the leading cause for the early aging of players. I could see this trend continuing as more and more teams implement shifts on players earlier and earlier in their careers.
While some changes in BB% and K% rates have factored into the recent change in hitter aging curves, earlier declines in BABIP and ISO are the leading causes. Players are no longer hitting with as much power. Additionally, the number of balls they put into play are turning more and more into outs. The increase use of better defensive alignments against a hitter earlier in his career may lead the player to age faster. It will be interesting to follow the trend in the future.
Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.
I wonder what those graphs would look like if we excluded Barry Bonds.
Not a whole lot, his is only 600 matched PA out of tens of thousands. I do wonder if PED use masked the earlier aging.
I think that’s been the biggest and clearest result of drug testing, even more than the overall drop in offense. The aging curve of players has advanced two or even three years over where it was prior to testing. You no longer see players excelling until they turn 40 like you did 10 years ago. Players these days are lucky to be able to excel into their late 30’s.
Yeah, no more Hank Aarons.
Do the new defensive alignments lead to players aging faster or getting unlucky faster.
Maybe it is better to say their production is declining sooner.
I still don’t really understand the methodology. When I look at players who even made the majors at age 21 between 2006-2013, I come up with a list of 31 extremely talented players. Age 22 there are 90 of a wider range of talent. Age 23 it becomes 175, etc.
Yet when I look at any of those individual players, they always seem to peak in, say, ISO, or WRC+, or whatever I look at, somewhere around age 28, the age traditionally thought of as the peak.
Here’s an experiment. Look at all 70 rookies in 2006, the year in question here. Find me *one* who peaked at age 21. Now explain how this is players only declining with age, and not sampling problems.
And I’m not trying to be an ass about it; if I am fundamentally misunderstanding this, I am trying to learn.
You’re not being an ass, I’m asking the same question. It seems that there are some biases here, the talent pool at the younger ages will primarily be made up of great talents, as you get older in age, there will be more players and their average talent level will be presumably less.
What kind of methodology is out there to approximate minor league production? If player X’s first year in the majors is 25 and that’s his best year, isn’t it possible that he improved while in the minors over his age 21-24 seasons? If player X’s double A numbers at age 22 can be translated into major league production it would go a long ways fixing the selection problems
In the 2nd article, Jeff ends the post with a graph AAA players age curve by wRC+. Not exactly what you are asking about but…
I have a methodology question too. Is this following the same player by age or all players averaged? If its all players then the numbers likely drop early because elite talent gets to the bigs at age 20, 21, by the time you get to 25 a lot of fringe players arrive. I think you are using player matched values. In that case, how many guys are in the 20 year old group vs the 25 year olds? I guess I don’t understand how you are generating these and I followed the links to the previous articles and its still unclear to me. It just doesn’t seem plausible that a 20 year old man is at peak athletic ability. It also doesn’t jive with the improvement that individual players make as the spend time in MLB (as iron mentioned).
“It just doesn’t seem plausible that a 20 year old man is at peak athletic ability.”
Why not? That’s about when we draft them into the army.
Of course, I’m sure they learn how to play smarter over time, so their peak production comes later.
20 year olds are not at peak muscle mass and strength that takes physical maturation. Players get better in their mid twenties in every sport, its not a coincidence.
If their peak production comes later, then there peak ISO would be later too.
Exactly. Who exactly are these alleged players that see, on average, a 10% decrease in production from in their age 28 season from their age 21 season?
There were 6 players in 2006 who were 21 or younger and had 100 PA. None of them had their best wRC+ that year. Only Delmon Young was close.
There 5 such players in 2007. Again, none of them had a their best wRC+ at the plate. Delmon Young actually got WORSE.
In 2008 there were 2 such players, Jay Bruce and Pablo Sandoval. And yet again, neither one of them had their peak season that year.
I can’t help but think something is going on here.
Yes, you’re right. It’s pretty clear the methodology is just wrong.
Yeah, I’m seeing a similar thing. The best players in baseball are, overall, still age twenty five to thirty, who also tend to have a few years of experience too. my anecdotal impressions of the best players in baseball aren’t data, but I feel like the combo of survivor bias and ignoring league wide changes is distorting the real changes.
Allright I randomly picked 30 rookie eligible players with qualified PA in 2006 and looked up their peak perfomrance measured by wRC+ since it takes league wide changes into account and park and all.
K.Morales: 26
H.Ramirez: 29
B.Zobrist: 28
J.Inglett: 30
C.Coste: 33
J.Hermida: 23
L.Milledge: 22
S.Costa: 24 (only played thru age 25)
R.Garko: 26
C.Quentin: 25
A.Ethier: 26
I.Kinsler: 26
B.Anderson: 26
C.Jackson: 26
N.Markakis: 24
K.Johjima: 30
M.Kemp: 26
B.Fahey: 25
R.Martin: 24
P.Fielder: 25
R.Spilborghs: 28
R.Shealy: 28
S.Drew: 23
H.Kendrick: 27
R.Zimmerman: 25
Mlk.Cabrera: 27
C.Duffy: 27
L.Scott: 28
D.Uggla: 30
A.Marte: 26
There is obviously a bias towards a younger age because a lot of these guys did not have carreers beyond age 27 or even 26 but nevertheless we see an
average peak age of 26.4
median peak age of 26
but nothing here suggests that players peak at 22 or so.
This is just the sort of list I did when the first article came out and I reached the same conclusions. If players are peaking at 21-22, you would find examples other than the rare Lastings Milledge flameout.
I’m not saying Jeff hasn’t found *something*, since the values have changed recently by the same methodology. But I think it is a difference in how teams are treating young and old players, not how actual individual players are aging. Or maybe it’s all just Mike Trout.
Don’t understand this at all either. The graph seems totally off and this idea that players peak at 21-22 simply doesnn’t seem true.
Wish Jeff would explain what the hells going on here and how he got those graphs.
Judging from the league values you provided, I suspect the K% and ISO curves would be significantly different if league adjusted statistics were used.
Even if not “significantly” different, without knowing the age distributions for each season, we don’t even know which way the bias is moving the curves…
No, players only get older, so the secular change would cause what we see.
So, the other issue is that each year (1) it’s a new year, and (2) every player is exactly one year older. This makes secular changes indistinguishable from faster aging.
I think Iron raises a good point. Obviously we are seeing some very good very young players these days – Trout, Machado – but not 18-year-olds. Power is very much in demand, so if ISO starts high shouldn’t we be seeing more young sluggers rushed to the bigs?
Start bunting on shifts and watch the numbers start to shift
yea, i really struggle with the methodology here as well… same reasons as Iron and others cite above.
i wanted to see if this passes the “smell test” and just took the 16 aged-30 qualifying batters in 2013 and looked at their aged 24 seasons as a proxy for their curves. as a caveat, i am not trying to pass this off as reasoned analysis – only a smell test.
the results were interesting: 13 / 16 players contributed higher wRC at 30 than 24 or 81% of my small sample – with 4 players not contributing meaningfully in 2007 (Freese, Nava, Venable and Choo.) The average difference in wRC was +29%.
The 3 players that saw a decline from 2007-2013 were Russel Martin, Yuni Escobar and JJ Hardy.
Your sample has, “survivor bias” issues. You didn’t sample all the 24 year olds that did not make it to 30 because they saw decreased performance.
You also can’t assume that all 24 year olds that don’t make it to 30 had decreased performance. Many never were any good to start with and thus didn’t make it.
wow another stupid article by fangraphs which tells nothing
To echo some of the other comments, Jeff you simply can’t just show is these graphs without putting them into league context. They become almost meaningless. And just showing us the league averages for 3 different years does not help much.
I mean you got all excited about the new (06-13) ISO curve, yet you show a league-wide decrease of 20 points from 06 to 13. The graph surely needs to be adjusted for league-wide changes. I don’t know what you decided to go from league-adjusted stats to non-league adjusted ones. That makes the issue a lot more confusing and nearly impossible to draw any conclusions.
That being said, even when adjusting for league averages, one has to be careful as to the sources(s) of any league-wide changes. For example, let’s say that there is a league-wide trend for K% among batters to go up (as there is).
Using the delta method, as you do, if that trend/change is caused by new players coming into the league who have higher K% or by the overall age of MLB players being higher, is some such “endogenous” reason, then you do NOT want to do a league-wide adjustment. If the increase in K% is caused by something having nothing to do with the batters themselves, such as umpire, strike zones, park changes, better pitchers, etc., such that these changes are expected to affect ALL batters equally, then you MUST adjust the stats in doing aging curves.
So, there are a lot of things going on here when league-wide changes are occurring that must be properly accounted for when doing aging curves.
Jeff, you have done great work with respect to aging and other things, but I am afraid that you are in a pitcher’s count right now as far as these issues are concerned.
I think the problem here is that Jeff is trying to discover the reasoning behind the league wide changes, and he’s postulated that a lot of it has to do with changes in aging curves. The methodology fails because he is trying to test two hypotheses in one.
Sorry for all the typos above. Ignore them.
One also has to be careful with the definition of “aging” with respect to aging curves (when using the delta method).
As Jeff explains withe respect to BABIP…
These types of aging curves that are not league-adjusted capture two things: One, an actual increase or decrease in various skills and observed performance from one age to another and two, overall changes in the league, external to the batters themselves (as I explained in my post above).
Those two things are exactly additive. One of the problems with that is that aging curves due to decline/increase in skills/performance are generally not linear (they usually look like a reverse log normal curve) while the curves due to league-wide changes are usually linear and constant (as long as the trend continues).
So, for example, let’s say that strikeouts typically decrease from age 20 to 17 and then increase from that point on. But let’s also say that over the time period of the curve, league-wide K rates have increased every year because of a gradually changing strike zone and better pitchers. Well, the new curve, after adding those two influences together (endogenous and exogenous) might make it appear as if a increase in K rate starts at 20 and then gets even steeper after age 27. So the shape of the “real” (skill-based) aging curve completely changes by adding a fixed value to the y axis, as you would do if something like the K rate increased every year for all players.
That may be what is happening in many of your curves, Jeff.
Using K%+, BB%+, ISO+, BABIP+, adjusting for league average and potentially park will solve the issue of changes in league average.
This, however, does not differentiate whether the change is due to the introduction of new players who are different from the previous league average, or general changes outside the batters that affect both the new players and the returning players.
K%- is the better way to phrase it
It would be interesting to see this research used to draw a bigger picture in terms of a projection model. While the current -0.5 WAR per year after age 30 seems to be commonly accepted, if you quantify your results, I wouldn’t be surprised to see a more accurate regression projection.
As I was reading this I couldn’t help but think Jeff is saying experience means nothing. I have little idea of delta or wrc+ or??? Except maybe Delta Dawn. Ha. Then I read the responses. I think Jeff does a great job. Maybe he missed on this one. I’m not completely certain but I think so. And I’ll give him this one.
By the way, why do you have to be so negative Grouchy? I don’t agree with a lot of stuff Cameron and others say here. But it is my favorite site. Try to be more positive. It ain’t all bad. In fact, it’s almost all good. Chill!
Why are they not adjusted for league averages or are they?.
Closer reading and it seems they are not. There have been significant changes in K rate, Hr/FB rates (which in part is due to the K rate increase so fewer FB’s) and a modest drop in BABIP regardless of age between the 2 periods. The changing and expanded strike zone could explain most of that (batters making weaker contact as a result of being on the defensive more often), as well as the shift which has reduced LHB’ers platoon advantage over RHB’ers (LHB’ers have traditionally been better represented among the top hitters).
Right, but could that be because batters are peaking earlier?
Staggering article, really interesting stuff.
It’s really not what you’d expect whatsoever nowadays, players at 21-22 arriving at their prime, and from then on it is all downhill. Totally agree that the improved minor league conditioning and training must be the reason for such an early peak, but I have another question.
Presumably PED’s is the sole reason for the early and steep decline hitters face nowadays, with late 20’s guys unable to turn around young phenoms heaters like the used to. So if we look long back in time, say 100 years (or however long it was, sorry I don’t know, Im English) before PEDS were prevalent, would the same early and rapid decline be shown?
Or is it because these young phenom arms are so good once they reach the show now, that only their peers (up to 27 say) can cope with them? Doesn’t explain why older (28+ guys) can’t hit their peers of the same age group though…
But the conclusion just isn’t true, there are countless examples of players who gain power as they hit 25-27. McCutchen leaps to mind immediately, but you can list almost any young player and see ISO increase with physical maturation.
Does this mean the likes of Rizzo aren’t going to experience the improvement that we all expect?
It is not clear to me how or if the rates were adjusted for the rate of players coming up at various ages. If Trout is the only player coming up at age 19, his age 19-20 change is valid. But if another lesser player comes up the subsequent year at 20 and the age 20-21 changes are averaged, a drop is going to appear despite a better year by Trout. One way to evaluate the effect is by comparing entry age cohorts. My suspicion is that for any cohort, aging has not changed as dramatically as suggested by the analysis thus far.
Maybe the graphs should follow players who entered at 21 for the whole career. You can’t introduce new players into the data at different ages because it changes the conclusions that can be drawn.
I’m a little disappointed Jeff zim never came back and addressed any of these questions.
I bet he will, but he’s probably waiting until he has a significant update rather than just some off-the-cuff rhetoric. I hope so!