March 15, 2013
The BP Wayback Machine
Do Spring Speeds Matter?
While looking toward the future with our comprehensive slate of current content, we'd also like to recognize our rich past by drawing upon our extensive (and mostly free) online archive of work dating back to 1997. In an effort to highlight the best of what's gone before, we'll be bringing you a weekly blast from BP's past, introducing or re-introducing you to some of the most informative and entertaining authors who have passed through our virtual halls. If you have fond recollections of a BP piece that you'd like to nominate for re-exposure to a wider audience, send us your suggestion.
Are velocity readings trustworthy this early in the year, or should we wait until Opening Day to start putting any stock in the speed of a pitcher's stuff? Mike examined the issue in the piece reprinted below, which originally published as a "Spinning Yarn" column on March 30, 2011.
Fastball speed in the major leagues is an important and oft-researched topic. As the 2011 season begins, the trickle of reports on pitchers’ fastball speeds that came out of spring training will turn into a flood of data. Some pitchers will be throwing a little faster than they were last year, while others will have lost a notch on their hard stuff.
Do these early fastball fluctuations matter, or does a pitcher who starts slow in March and early April regain the life on his fastball later in the year? Pitchers like Joel Pineiro and Mat Latos who have lost velocity this spring seem like candidates for injury or performance drop-offs, but maybe we should start paying attention to how pitchers perform only once their teams have headed north. Does what happens in Arizona stay in Arizona? Is fastball speed in April, when the games count and the opposing hitters are all of major-league quality, what really matters?
Let’s look at how average fastball speed has changed throughout the year, as compared to a pitcher’s typical speed as measured by PITCHf/x over the 2008-2010 seasons.
Fastball speed for an average major-league pitcher starts at its lowest point in early April, rises by about 1.0-1.5 mph to a peak in the month of July, and declines gradually thereafter. These trends apply similarly to starting pitchers and relief pitchers. However, the fastball speed curve looks suspiciously like a graph of temperature.
Fastball speed has a strong dependence on temperature, irrespective of the point in the season.
If we approximate the relationship between temperature and fastball speed as linear, we find that average fastball speed increases by about one mph for every 37-degree increase in game-time temperature.
Adjusting for temperature removes some, but not all, of the seasonal pattern.
Interestingly, the 2008 season still shows a noticeable increase of nearly one mph from April to July, whereas the 2009 and 2010 seasons have much flatter fastball speed profiles once temperature effects are removed. The origin of the differences in profiles between seasons is unknown.
Some PITCHf/x fastball speed data hails from outside the regular season. The limited amount of such data makes a date-based comparison too noisy to be useful, but the overall temperature-adjusted fastball speed as compared to the regular season average is as follows:
There are some challenges in applying these league-wide results to individual pitchers. Foremost among them is PITCHf/x camera calibration, which varies over time and from park to park. My method for correcting pitch location data to account for camera calibration problems was detailed here, and I used a similar method to correct fastball speed data.
The extent to which spring training fastball speeds are predictive of regular season fastball speeds is of particular interest. PITCHf/x data is collected from two spring training sites, Surprise and Peoria, Arizona, which play home to the Texas Rangers, Kansas City Royals, San Diego Padres, and Seattle Mariners. PITCHf/x information from spring training is primarily available for those four clubs and their visiting Cactus League opponents. Thus, PITCHf/x data from Grapefruit League teams is mostly nonexistent. Another smattering of spring training PITCHf/x data is available from the handful of exhibition games played in major-league ballparks prior to opening day.
If we look at fastball speed data collected in spring training last season, we can compare the speeds for individual pitchers to their average fastball speeds in September 2009. Then we can see how well this difference predicted the individual pitcher’s fastball speed for the 2010 regular season.
There is definitely a predictive relationship between the fastball speed change observed in spring training and the fastball speed change observed in the regular season. Pitchers on average retained 41% of their spring training speed changes into the regular season, and the correlation coefficient for the sample here is r=0.53.
The pitchers with the biggest changes in fastball speed from September 2009 to spring training 2010 are as follows.
Many, but not all, of the spring training fastball speed changes presaged a similar change in the regular season. Some of them, of course, have relatively harmless explanations. For example, C.J. Wilson lost three mph while transitioning from the bullpen to the starting rotation.
Since spring training data collection is so limited, perhaps early regular-season data, which is collected for every team and every pitcher, would be a better predictor of ultimate fastball speed across the full upcoming season. Let’s compare the fastball speed changes observed in the first two weeks of the 2010 regular season as compared to September 2009 to the fastball speed changes observed in the remainder of the 2010 season as compared to September 2009.
The correlation does appear to be a little stronger here, with pitchers on average retaining 52% of their early April speed changes, with a correlation coefficient of r=0.66. Let’s look at some of the individual pitchers.
There are some players with real velocity changes on the list above, but there are also some examples that reveal the shortcomings of a blind reliance on this method while using an automated pitch classification system. Valverde is the obvious problem here, as his fastball speed didn’t really change much from 2009 to 2010. Instead, the Major League Baseball Advanced Media (MLBAM) pitch classifier had trouble separating his fastballs from his splitters in the early part of 2010. Batista and Hughes both greatly increased their cut fastball usage from September 2009 to April 2010, and since I included cutters when calculating average fastball speed, their speeds show a decrease due to changing pitch mixes.
As spring training draws to a close in 2011, which pitchers have displayed the greatest fastball speed changes this year among those Cactus League pitchers for whom there is PITCHf/x data?
Feliz’s fastball speed presumably dropped because of his latest experiment with starting. Some of the other changes may also have innocent explanations upon further investigation. Moreover, this list may not capture some of the subtler indications of trouble hidden in the fastball speed data.
For example, Harry Pavlidis identified a troubling decrease in Latos’ fastball speed during spring training, but his overall average speed change does not stand out as much as the decrease over time. Latos had an average fastball speed of 93.0 mph in September 2010. In spring training this year, his four PITCHf/x-recorded starts had average fastball speeds of 93.7, 93.4, 91.6, and 91.6 mph, after adjusting for temperature and camera calibration. Now comes news that Latos has inflammation in his right shoulder.
Changes in fastball speed offer promise as a diagnostic and predictive tool, but there are pitfalls to avoid, and our techniques for separating the wheat from the chaff must be improved. It will be interesting to see whether pitchers such as Lilly, Pineiro, and Latos recover their lost velocity and go on to successful seasons or whether they will have to work against injury and slower fastball speeds in 2011.
Methodological improvements are needed to adjust for changes in pitch mix between fastball types and to account more carefully for atmospheric changes that affect the drag coefficient of the ball. Nonetheless, there appears to be substantial value in the current form of this data.