Starter performance trends by fastball velocity

The year of the pitcher is now in its third year, and it feels safe to say that the offensive boom of the late 1990s and early 2000s is behind us. That is no secret. Two of the potential causes of the depression, the regulation of performance enhancers and the increased emphasis on defense, have been well-discussed. However, those are broad factors. I was curious to know if the evolution of the game has helped all types of pitchers, equally. Specifically, I decided to look at starter performances based on their fastball velocities.

There have been several high-profile pitchers who have lost some of the heat off their fastballs from previous seasons, though with mixed results. Felix Hernandez is down 1.4 miles per hour from 93.4 in 2011 to 92.0 in 2012 per PITCHf/x, but his 9.50 K/9 and his 2.37 BB/9 are career bests. On the other hand, Tim Lincecum is down 1.8 miles per hour from 92.2 in 2011 to 90.4 in 2012, and his 9.89 K/9 is below his peak performance from 2008-2009, while his 4.39 BB/9 is a career worst.

A look at the FIP leaderboard, shows it littered with slow tossers like Jered Weaver, Ryan Dempster, and Kyle Lohse. So, is it easier now than it was a few years ago to succeed with a slow fastball? Perhaps it is the offensive depression and not guile that has sheltered Hernandez, a possibility that would be particularly damning for Lincecum.

To find out, I decided to group starters into three tiers based on their fastball velocities and measure the effectiveness of those tiers over the past six seasons. Then, I could compare the trends of those results to Hernandez, Lincecum, and other starters of interest, especially those who dropped from one tier to another over the time frame.

First, I needed to establish the boundaries of the tiers. My main goal was to set the boundaries at nice round numbers for ease of comprehension, but I also wanted the tiers to be of a similar size and contain enough pitchers to support a statistical conclusion. I settled on a Tier 1 for starters who threw fastballs that averaged more than 92 mph, a Tier 3 for starters who threw 90 mph or slower, and a Tier 2 that fell between the other two tiers. I treated each season independently, and I limited the scope of the research to pitchers who exceeded the innings qualification.

Here are the number of starters who fell into each tier over the past six seasons:

             2007    2008    2009    2010    2011    2012  Total
Tier 1        20      25      26      30      36      35     172
Tier 2        24      32      29      34      26      36     181
Tier 3        35      29      20      27      28      28     167

Without a deeper look at their statistics, the tiering of starters already shows an interesting trend. Combined, the three tiers have a similar number of pitchers. However, the distribution of the tiers has changed significantly in recent seasons. In 2007, only 20 starters threw above 92 mph. Halfway into 2012, 35 have. In contrast, 35 starters threw 90 mph or softer in 2007, and fewer than 30 have done the same in every season since.

The easy conclusion to draw is that the increase in hard-throwing starters has been responsible for the improved numbers of pitchers over the last few seasons, but I found that was not the case. In fact, Tier 1 starters have combined for a 2012 FIP worse than the previous five seasons, including 2007 when offense was still king.

FIP for Tier 1 (92+ mph Fastball velocity), Tier 2 (90-92 mph Fastball velocity), and Tier 3 (90- mph Fastball velocity) starters from 2007-2012:


Unexpectedly, it is the Tier 3 starters who have continued to improve in 2012 while Tier 1 and Tier 2 starters have both performed worse than last season. Why is that the case? FIP measures pitcher performance by three factors that are within his control: strikeouts, walks, and home runs allowed. I tracked K/9, BB/9, and HR/9 for the starter tiers over the same window to try to identify the reason for the success of Tier 3 starters.

Strikeouts per nine innings for Tier 1, Tier 2, and Tier 3 starters from 2007-2012:


Tier 3 starters have seen an improvement of close to a full strikeout per nine since 2007, but Tier 1 and Tier 2 starters have seen similar increases. Tier 1 starters still maintain more than a 1.0 K/9 lead over Tier 3 starters. Strikeouts are evidence for the improvement of pitchers over hitters, in general, but not of the relative gains made by Tier 3 starters.

Walks per nine innings for Tier 1, Tier 2, and Tier 3 starters from 2007-2012:


Walks are where Tier 3 starters have made the most headway. In 2007, Tier 1 and Tier 3 starters had nearly identical walk rates of 2.89 BB/9 and 2.82 BB/9, respectively. Since, the two have diverged. Tier 1 starters are just shy of 3.0 BB/9 so far in 2012 while Tier 3 starters have cut their walks to just 2.53 BB/9.

Home runs per nine innings for Tier 1, Tier 2, and Tier 3 starters from 2007-2012:


Counter-intuitively, Tier 1 and Tier 2 starters have allowed close to 1.00 HR/9 so far in 2012, after sitting close to 0.80 HR/9 in 2010 and 2011. On the other hand, Tier 3 starters have continued their steady decline in home runs allowed in 2012. For the first time in the last six seasons, Tier 3 starters are allowing less than 1.00 HR/9, and they are allowing fewer home runs than either Tier 1 or Tier 2 starters for the first time, as well.

Homestretch: The 1967 AL Pennant Race, Part 3
A tight race shows no signs of letting up.

All told, the limiting of walks and home runs has allowed Tier 3 starters to improve more than their counterparts. As I mentioned previously, the career-best 2.37 BB/9 for Felix Hernandez shows that he has followed the trend of other starters with non-elite velocity, despite only recently becoming one of them. In addition, Hernandez has allowed 0.61 HR/9, down from his career average of 0.74. On the other hand, Lincecum is walking more batters, 4.39 BB/9, and allowing more home runs, 0.95 HR/9, than he ever has.

While it is true that the tier boundaries I set are somewhat arbitrary, I find it interesting that Hernandez and Lincecum could each drop from Tier 1 in 2011 to Tier 2 in 2012 and yet show such different results. I wanted to know whether one or the other was an unusual case, and so I decided to look at other starters from 2007-2012 that fell from one tier in one year (y) to another in the next (y+1) to see if they had a similar inability to adjust as Lincecum.

To measure the differences in performance of tier droppers, I looked at both ERA and FIP. Lincecum’s 2012 ERA is more than 2 runs higher than his FIP, and I was curious if velocity loss somehow correlated to performances that were worse than the peripherals would suggest. In addition, I looked at the percentage of fastballs thrown of all pitches thrown, represented by FA%. Intuition told me that the success of pitchers like Hernandez might be because of an adjustment to their pitch selections.

Tier droppers from 2007-2012:

Tier in y  Tier in y+1  ERA in y ERA in y+1   FIP in y   FIP in y+1   FA% in y  FA% in y+1
     1          2       3.32       3.85         3.63       3.72        43.1%      34.30%
     2          3       4.11       3.97         4.13       4.18        38.3%      29.20%
1 or 2     2 or 3       3.70       3.91         3.87       3.94        40.8%      31.80%

In the window, there were 16 starters who fell from Tier 1 in one season to Tier 2 in the next, including Lincecum and Hernandez. There were 15 starters who fell from Tier 2 in one season to Tier 3 in the next. Both sets showed remarkably similar numbers before and after their loss of velocity. Combined, the 31 starters increased in FIP from 3.87 to just 3.94. I expect that consistency is due, in part, to an increased dependence on their secondary pitches. The average tier dropper used his fastball close to 10% less often the year after losing velocity.

It seems that a loss of velocity has not prevented the success of starters in recent seasons, and tier droppers do not show uncharacteristic performance levels for their new tiers compared to pitchers who have maintained velocities or those that were previously unqualified by innings. Most have made an adjustment in their pitch selections to maintain their success, and so it should be no surprise that Hernandez, who has quality secondary pitches, continues to perform well having done the same. So what to make of Lincecum?

Despite seeing their FIP increase from 3.63 to just 3.72, fallers from Tier 1 to Tier 2 saw a much greater jump in their ERA, after losing velocity, from 3.32 to 3.85. Lincecum is the most powerful example. His ERA has climbed from 2.74 in 2011 to 5.93 so far this season, while his FIP has increased from 3.17 to just 3.72.

This season, Ricky Romero, Josh Beckett, and Justin Masterson follow the same pattern, with a much greater increase in their ERA than their FIP from 2011. In the sample I chose, that is unusual. The closest precedent, oddly enough, is Tim Lincecum, who fell from Tier 1 to Tier 2 in 2009-2010, before he recovered some velocity in 2011 and lost it again.

With only a half-season of work, chances are that those discrepancies in ERA and FIP will decrease in the second half. Even if Lincecum pitches in the second half the way he did in the first, I would expect his ERA to fall. And recent history of pitchers with similar loss of velocity suggests that Lincecum has a chance to perform better.

However, there are some specifics for Lincecum that have me worried. Compared to 2011, Lincecum has barely decreased his fastball usage from 42.1% to 40.4%, which is much less of a drop than that of the average starter who loses velocity. Instead, he has used fewer sliders—24.1% to 18.4%—and more curves—6.4% to 11.2%. It makes me wonder if he could be injured. Last season, his slider was his most effective pitch by PITCHf/x Pitch Value. This season, it is his least effective, a signal, I fear, that he could buck the positive trend.

References & Resources
Statistics from FanGraphs.

Print This Post
Sort by:   newest | oldest | most voted

Cool stuff, thanks.

My stats education is about 25 years old but isn’t it more effective to try to plot velocity as a continuous variable first and then cohort them if the data shows it is appropriate? I think that would make sure you use the right inflection points and avoid overlap of groups as opposed to arbitrarily applying limits to each cohort.

Scott Spratt
Scott Spratt

Hey, Mike.  I think that is a fair criticism.  If you exclude outliers like RA Dickey, fastball velocity is normally distributed, or close to it.  I definitely could have been more sophisticated in establishing boundaries, but I feel that the arbitrary groupings I created on either side of 90 and 92 mph are simple and memorable, and the resulting tiers are large enough and similar in size so the risk of a skewed result are not too great.

Just numbers
Just numbers

The Tiers are deceptive.  92+ is a large range, with differing physical requirements – each MPH requires more and more energy.  Tier2 is rather small in comparison, with nearly balanced requirements.  Tier3 is even more wide open.

You need some backup or at least some presentation for ‘equal energy tiering’.  In case you have forgotten, kinetic energy goes as the velocity squared.  You’re not going to get ‘large enough tiers’.  Tough, but I cannot take this tiering seriously.