Touching BasesFebruary 18, 2010
The Verducci Effect
By Jeremy Greenhouse

On Monday, Will Carroll noted that the Verducci Effect was being discussed on MLB Network. On Tuesday, Tom Verducci posted his ten young pitchers at risk of the Effect. Then to top it off, yesterday Josh Hermsmeyer unveiled a free player injury database. I've been meaning to research the Verducci Effect for some time, so this seemed like as good a time as any.

The Verducci Effect, also known as the Year-After Effect, is defined by BP as "a negative forward indicator for pitcher workload," Specifically, pitchers under the age of 25 who have 30-inning increases year over year are at risk. David Gassko's research pointed to the opposite. With pitch by pitch data from FanGraphs and disabled list data from Rotobase, I attempt to expand on Gassko's preliminary analysis, although purely numerical research on injury prediction and pitch limits will never come close to showing conclusive results.

I found 340 pitchers who pitched three consecutive years in MLB at ages 25 and under since 2002. 140 of them fit the Verducci Effect, while 200 did not. Here's the data.

Verducci Group

Year One 48.3 90.4 7.3 3.7 41.5% 0.313 20.6% 51.4% 18.71% 51.6
Year Two 126.5 91.1 7.0 3.4 44.2% 0.302 19.7% 52.0% 20.14% 36.6
Year Three 113.1 90.9 7.2 3.3 43.3% 0.305 19.6% 51.4% 29.50% 59.4
Difference -13.3 -0.2 0.2 -0.1 -0.8% 0.003 -0.1% -0.6% 9.35% 22.8

Non-Verducci Group

Year One 80.5 91.3 7.0 3.6 42.1% 0.301 20.6% 52.0% 20.49% 48.9
Year Two 65.7 91.3 7.1 3.6 42.2% 0.326 20.2% 51.4% 34.63% 69.5
Year Three 85.2 91.5 7.3 3.5 41.9% 0.306 20.7% 51.0% 36.10% 57.6
Difference 19.5 0.1 0.2 -0.1 -0.3% -0.019 0.5% -0.4% 1.46% -11.9

The first point of interest is the decrease in innings pitched for those under the influence of the Verducci Effect. I should preface the rest of this analysis with a few popular credos: TINSTAAPP, regression to the mean, and small sample size. First, pitching is an inherently risky business. Dave Cameron recently wrote a great piece on how successful young pitchers often peak early. This problem is exacerbated by the nature of the Verducci Effect, which dictates that pitchers establish a career high in innings pitched. If you take any group of players who establish a career high in any category, chances are that they will regress to the mean the following year. Finally, my sample again only contains 140 Verducci pitchers. One can't draw important conclusions from a sample of that size. You've been given fair warning.

In general, 25-and-under pitchers improve their peripherals in their third year. Their strikeout rate trends up while their walk rate trends down. Gassko found similar results. I'm not so interested in whether or not young pitchers improve; I'm looking to see where Verducci Effected pitchers differ from other pitchers.

Therefore, the Difference row is the row of interest, as it represents the change from the innings-jump year to the Year After. There are four terms in the Difference row that report different positive/negative signs (besides innings pitched) between each group. BABIP, velocity, whiff rate, and days per DL trip. That Verducci Effected pitchers suffer worse luck based on BABIP and that their counterparts exhibit better fortune speaks to the infallibility of regressing to the mean. I'm not so interested in the contact rate of pitchers, but I decided to further explore the possible velocity and injury aspects of the Verducci Effect. So I turned to the statistical technique of regression analysis.

First, I tried predicting fastball velocity using several separate variables for age, past velocity, and past workload. I've looked at the topic of velocity curves before. Velocity generally peaks during a pitcher's mid twenties. Here are the regression results, which I've broken down by variable type.

Age Coefficient P-Value
32 Up NA NA
29-32 0.14 0.03
26-29 0.30 0.00
26 Down 0.55 0.00

Younger pitchers have a .5 MPH advantage over older pitchers in velocity.

Velocity Coefficient P-Value
Year One 0.18 0.00
Year Two 0.78 0.00

Fastball velocity from the previous year has nearly five times as much predictive value as fastball velocity from two years ago.

Workload Coefficient P-Value
Year One Pitches (1000) 0.06 0.11
Year Two Pitches (1000) -0.14 0.00
Verducci Effect -0.30 0.02

The previous year's workload helps predict velocity. Throwing a thousand pitches in a year coincides with a drop in velocity of more than a tenth of a mile per hour. This could represent the difference between starters and relievers, in that starters throw more pitches at a lower velocity than relievers. Also, pitchers who have undergone the Verducci Effect have thrown softer than non-Effected pitchers to the tune of 0.3 MPH.

Next, I ran another linear regression to predict days spent on the disabled list in a pitcher's third consecutive year of pitching.

DL History Coefficient P-Value
Year One DL Trips 4.19 0.09
Year Two DL Trips 5.90 0.01
Year One DL Days -0.03 0.46
Year Two DL Days 0.18 0.00

First off, predicting future health is hard. While I was able to predict nearly 90% of a pitcher's fastball velocity without developing a very sophisticated model. The disabled list model explains only 6% of a pitcher's health. Nevertheless, injuries from the previous year are significant, as each trip to the DL tends to yield another several days on the DL the following year.

Age Coefficient P-Value
32 Up NA NA
29-32 -1.31 0.60
26-29 -3.72 0.13
26 Down -0.16 0.96

Age isn't a very strong predictor of future injuries. Pitchers on either extreme of the age spectrum are most at risk, but the results aren't significant. Verducci might've chosen a wise cutoff at age 25, as this table shows that there could well be a point at which pitchers grow less vulnerable.

Workload Coefficient P-Value
Year One Pitches (1000) 0.6 0.78
Year Two Pitches (1000) 4.3 0.06
Verducci Effect 0.63 0.90

The Verducci Effect, like most everything else I tested, is not significant in predicting future injuries. Injuries are hard enough to predict as is, and there's certainly no straightforward rule of thumb. A high workload does coincide with a trip to the DL the following year, though the causative effect may be that pitchers who throw a lot of pitches have more opportunities to get injured, rather than the pitches placing more stress on their arms.

Verducci identifies the likes of Felix Hernandez and Josh Johnson as pitchers at risk. Verducci Effect or not, those guys aren't going to replicate their spectacular seasons. But Verducci also points to lesser pitchers such as Homer Bailey and Joba Chamberlain, who failed to live up to their prodigious potential last year. Bailey's fastball velocity leaped up three MPH last year while Joba's velocity dipped by a similar amount. I say if they stay healthy, they both improve on their performance from last year, but chances are at least one of them hits the DL. The data show that workload and age help predict production, velocity, and injuries, but the jury's still out as to whether the Verducci Effect helps explain the nexus between injury and risk beyond what one would expect from young pitchers with taxing workloads.


"The Verducci Effect, like most everything else I tested, is not significant in predicting future injuries."

Isn't this just a polite way of saying, "there is no Verducci Effect"?

From what I see, the VE went from "intriguing hypothesis" to "universally-accepted metaphysical truth" in the span of about 5 minutes, skipping over the step in which smart people who understand statistics confirm its scientific validity. I'm glad to see the air being let out of it.

Jeremy, nice work trying to get at formalizing this effect. We need to do more of this to investigate the causal nature of those innings pitched.

Quick request: Could you modify the first two tables to also show us the variance in each cell? You have the means and the difference between year 2 & 3, I think adding the variance would help paint a clearer picture.

For example, the mean innings pitched in year 3 drops by 13 relative to year 2. So we can see that on average the group saw reduced action, but if the variance was included, we could see if there was a large divergence in the response to those extra innings pitched.

Thank you for the post.

This analysis seems to treat all Verducci Effect pitchers equally but intuitively it would make sense that pitchers with bigger innings increases would be riskier than pitchers who barely qualified for the Verducci group. Is it possible to break the data into subsets to see if there is a more pronounced effect at higher increases?

To BD,
The reason why he says it isn't statistically significant is because there could be partially true but below the level of noise. It's not fully disproved but considered extremely likely to be correct.

I'd like to see if there is any difference between power pitchers and finesse pitchers. If only we had more information to go on

OR, while it isn't "fully disproved," Verducci and Carroll haven't yet proven it either. It's their job to prove there is an effect. They have not done so, but treat it as if it is proven.

I honestly don't know how you could call Cameron's piece "great" unless you didn't read it. It's short on evidence and shoddily reasoned.

This is a well-researched piece, but any legitimate analysis of this effect needs to include *all* innings, not just Major League innings. If a pitcher throws 100 innings in the minors and 100 innings in the majors, this study will treat him as "100 innings" for the season, when in fact, that's not the case.

It's hard to get accurate historical data on the minor leagues, but investigating players under the age of 25 without that data is by definition mischaracterizing many of the player seasons in the data set.

I will second the fact you have to include minor league innings or the results don't mean anything.

I'm not saying the "Verducci Effect" is necessarily true, but all you looked at was the effect on injuries, another aspect Tom Verducci looks at is a decline in performance, which I think makes the results a bit more conclusive...he doesn't simply look at injury risk.

Sorry, my mistake, you did at least address velocity, but still, my point was Verducci doesn't simply narrow the effect he looks at to just potential injuries.