How does ‘Stuff’ Impact Performance for Middlebury College’s Pitching Staff?
Ethan Gavin
May 19th, 2023
library(tidyverse)
library(broom)
library(moderndive)
library(kableExtra)
library(openintro)
library(ggfortify)
library(sjPlot)
library(sjmisc)
library(sjlabelled)
library(ggplot2)
library(knitr)
library(png)library(readr)
MiddleburyTrackManPitchingData <- read_csv("MiddleburyTrackManPitchingData.csv")
MiddPitch <- read_csv("MiddPitch.csv")
pitching_joined <- MiddPitch %>%
inner_join(MiddleburyTrackManPitchingData, by = "Pitcher")Introduction
The Middlebury College baseball team has been very successful over the past two seasons, going to the NESCAC Championship game both times and winning one. One of the reasons the team has seen so much success on the field recently is through its pitching staff. The recent success results from the current pitching staff’s possession of better stuff and command of pitches than in previous years.
In my position as the Director of Baseball Operations for the Middlebury College baseball team, I have collected a significant amount of pitching data from our pitching staff over the past two seasons, and even more so this past year with our addition of a TrackMan pitching device. Our wealth of data collection devices and in-game charting has provided me with data I have used to help our pitching staff improve.
This research project assesses what pitching factors contribute to the most success for our pitchers this season. My analysis is split into two parts. The first is how a pitcher’s entire arsenal impacts their overall performance, while the second is how individual pitch-type attributes contribute to the performance of the pitch.
To assess how a pitcher’s entire arsenal impacts their overall performance, I examined how different attributes of the pitcher’s arsenal, including average fastball velocity, command, and CSW% impact in-game performance statistics. The two statistics I used to assess performance are Fielding Independent Pitching (FIP) and Weighted On-Base Average (wOBA). FIP is a good statistic to measure the performance of our pitchers because it “focuses solely on the events a pitcher has the most control over – strikeouts, unintentional walks, hit-by-pitches, and home runs” (MLB Definition of FIP). Because there is significant variability in baseball gameplay, FIP removes some of the unlucky factors that may result in a pitcher performing worse than expected. Furthermore, wOBA is a strong candidate to assess the performance of Middlebury’s pitching staff because it more heavily weights extra-base hits than singles and walks/hit batters. Pitchers who give up more extra-base hits typically give up harder-hit baseballs and more runs, which indicates poor pitching performance. To read more about the MLB weights for wOBA, visit this website: (MLB wOBA).
To assess how individual pitch-type attributes contribute to the performance of the pitch, I used both the game chart statistical data and our TrackMan data for pitch metrics like horizontal and vertical break and spin rate. Similarly to the comparison with the pitch arsenals, I utilized wOBA for the individual pitch types to assess the performance of the pitch and used regression analyses with pitch metrics and the pitch wOBA to better understand how pitch factors contribute to the pitch performance.
Arsenal Stuff and Performance
MiddPitch %>%
select(Pitcher, ERA, FIP, WRKK, CSW, wOBA) %>%
arrange(Pitcher) %>%
kbl(col.names=c("Pitcher", "ERA", "FIP", "WR2K%", "CSW%", "wOBA")) %>%
kable_styling()| Pitcher | ERA | FIP | WR2K% | CSW% | wOBA |
|---|---|---|---|---|---|
| Atwood, Jackson | 5.11 | 3.47 | 0.60 | 0.28 | 0.336 |
| Crider, Cole | 7.63 | 4.98 | 0.57 | 0.26 | 0.403 |
| Dessart, Spencer | 3.55 | 3.88 | 0.65 | 0.30 | 0.319 |
| Duarte, Sawyer | 7.96 | 6.41 | 0.56 | 0.25 | 0.412 |
| Gatland, Andrew | 6.14 | 4.88 | 0.59 | 0.29 | 0.381 |
| Gattuso, Gavin | 135.00 | 31.20 | 0.60 | 0.21 | 0.777 |
| Gilmartin, Zander | 6.52 | 7.41 | 0.52 | 0.24 | 0.393 |
| Gustavson, Henry | 2.84 | 4.12 | 0.65 | 0.26 | 0.303 |
| Handa, Kunal | 4.63 | 4.29 | 0.59 | 0.28 | 0.303 |
| Knightly, Dylan | 3.52 | 6.03 | 0.75 | 0.29 | 0.391 |
| Lessing, Justin | 2.87 | 3.69 | 0.68 | 0.34 | 0.299 |
| Mosier, Freddy | 3.57 | 5.11 | 0.64 | 0.29 | 0.327 |
| Price, Alex | 4.20 | 4.04 | 0.62 | 0.34 | 0.322 |
| Ritch, Alec | 4.91 | 8.84 | 0.55 | 0.15 | 0.421 |
| Rosario, Alex | 9.58 | 7.49 | 0.46 | 0.25 | 0.332 |
| Tross, Owen | 2.13 | 4.99 | 0.56 | 0.28 | 0.305 |
Avg. Fastball Velocity vs. FIP and wOBA
The first analysis I used to determine how a pitcher’s arsenal impacts their performance involves their average fastball velocity. One of the most discussed components of a pitcher’s arsenal is the speed at which they throw, so understanding how fastball velocity impacts overall performance is crucial to my analysis.
MiddPitch %>%
ggplot(aes(x=avg_FB_velo, y=FIP)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("Average Fastball Velocity")+
ylab("FIP")+
ggtitle("Average Fastball Velocity (mph) vs. FIP")
\[ \hat{FIP}= 145.17 - (1.64)(AvgFBVelo) \]
fbfip <- lm(FIP ~ avg_FB_velo, data=MiddPitch)
tab_model(fbfip)| FIP | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 145.17 | 61.61 – 228.73 | 0.002 |
| avg FB velo | -1.64 | -2.62 – -0.65 | 0.003 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.474 / 0.436 | ||
Based on my regression analysis, changes in average fastball velocity do impact FIP. Regardless of the sample of pitchers taken, pitchers at the Division 3 level should experience better FIP with increases in average fastball velocity. Despite many attributes being a part of a pitcher’s arsenal, including command of pitches and quality of off-speed pitches, average fastball velocity does play a significant role in determining a pitcher’s success.
MiddPitch %>%
ggplot(aes(x=avg_FB_velo, y=wOBA)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("Average Fastball Velocity")+
ylab("wOBA")+
ggtitle("Average Fastball Velocity (mph) vs. wOBA")
\[ \hat{wOBA}= 2.77 - (0.03)(avgFBvelo) \]
fbwOBA <- lm(wOBA ~ avg_FB_velo, data=MiddPitch)
tab_model(fbwOBA)| w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 2.77 | 1.32 – 4.21 | 0.001 |
| avg FB velo | -0.03 | -0.05 – -0.01 | 0.003 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.472 / 0.434 | ||
Similarly to average fastball velocity and FIP, average fastball velocity does impact wOBA. The regression analyses for average fastball velocity with FIP and wOBA show the importance of pitchers throwing faster. The results of these regression analyses are not a surprise because throwing harder gives pitchers a larger room for error. The less time hitters have to react, the harder it is to square the pitch up. Furthermore, the harder pitchers throw their fastballs, their offspeed pitches tend to have a higher average velocity, which makes the pitch much tougher to hit. Pitchers should spend a significant time in the offseason working to increase their fastball velocities.
WR2K% vs. FIP and wOBA
Examining how command impacts performance is important to determine what factors impact pitching performance the most. One of the metrics used to assess pitching command is win the race to two strikes percent (WR2K%). A pitcher wins the race to two strikes if they reach two strikes in the count before throwing two balls. Generally speaking, the pitcher being ahead in counts puts them in better positions to succeed because hitters are forced to be more defensive to prevent them from striking out.
MiddPitch %>%
ggplot(aes(x=WRKK, y=FIP)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("WR2K%")+
ylab("FIP")+
ggtitle("WR2K% vs. FIP")
wr2k_reg <- lm(FIP ~ WRKK, data=MiddPitch)
tab_model(wr2k_reg) | FIP | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 14.14 | -19.60 – 47.87 | 0.384 |
| WRKK | -12.03 | -67.98 – 43.92 | 0.652 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.015 / -0.055 | ||
My regression analysis of FIP and WR2K% shows that changes in WR2k% do not have a significant impact on FIP. While, intuitively, pitchers want to get ahead of hitters in the count, being in those positions does not necessarily lead to pitchers getting outs. Hitters are willing to get behind in counts to pitchers that do not have good stuff because they know they will not be overwhelmed by what the pitcher throws. The lack of a relationship between WR2K% and FIP shows the importance of command coupled with stuff.
MiddPitch%>%
ggplot(aes(x=WRKK, y=wOBA)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("WR2K%")+
ylab("wOBA")+
ggtitle("WR2K% vs. wOBA")
wr2k_wOBA <- lm(wOBA ~ WRKK, data=MiddPitch)
tab_model(wr2k_wOBA) | w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 0.46 | -0.13 – 1.05 | 0.114 |
| WRKK | -0.14 | -1.11 – 0.83 | 0.762 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.007 / -0.064 | ||
Similarly, there is no statistical significance seen between WR2K% and wOBA. This finding echoes the importance of pitchers having good stuff. While having good command is certainly a crucial part of pitching, the combination of it with good stuff can lead to success on the mound.
CSW vs. FIP and wOBA
MiddPitch%>%
ggplot(aes(x=CSW, y=FIP)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("CSW%")+
ylab("FIP")+
ggtitle("CSW% vs. FIP")
\[ \hat{FIP} = 26.65 - 73.21(CSW) \]
csw_FIP <- lm(FIP ~ CSW, data=MiddPitch)
tab_model(csw_FIP) | FIP | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 26.65 | 7.26 – 46.03 | 0.011 |
| CSW | -73.21 | -144.21 – -2.20 | 0.044 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.259 / 0.206 | ||
There is statistical significance seen for Called Strikes + Whiff Rate percent (CSW%) and FIP. Because CSW% is heavily based on stuff through its focus on swings and misses, its significance with FIP shows the importance of stuff in performance. Furthermore, the confidence interval shows that changes in the sample will still show that increases in CSW% will result in decreases in FIP, and depending on the sample, the pitchers will see more substantial decreases in FIP with increases in CSW%.
MiddPitch%>%
ggplot(aes(x=CSW, y=wOBA)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("CSW%")+
ylab("wOBA")+
ggtitle("CSW% vs. wOBA")
\[
\hat{wOBA} = 0.73 - (1.33)(CSW)
\]
csw_wOBA <- lm(wOBA ~ CSW, data=MiddPitch)
tab_model(csw_wOBA) | w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 0.73 | 0.40 – 1.06 | <0.001 |
| CSW | -1.33 | -2.54 – -0.12 | 0.033 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.285 / 0.233 | ||
Similarly, there is statistical significance seen between changes in CSW% and wOBA. These findings make sense because pitchers are less likely to give up hard-hit balls or unlucky hits if hitters are swinging and missing and taking pitches more. These findings stress the importance of the development of stuff in pitchers because my regression models all favor stuff over command when assessing performance. While command can make a pitcher with good stuff elite, it seems command on its own is not enough to see consistent success on the mound.
Individual Pitch Type Performance
To assess how stuff impacts pitch performance, I examined how metrics like average velocity, spin rate, and horizontal and vertical break impact wOBA (weighted on-base average) for the individual pitch type. When assessing how stuff impacts pitch performance, it is important to be cognizant of how the pitch is executed (location and sequencing) and how it fits with the rest of the pitcher’s pitch mix. However, I focused on the TrackMan pitch metrics to understand pitch performance.
Fastball Stuff vs. Performance
My first analysis was of fastballs. Because pitchers throw different types of fastballs (e.g., 4-seam fastballs and sinkers), I did not do a regression analysis examining how vertical and horizontal break impact the performance of the pitch. I instead focused on average velocity and spin rate to determine how stuff impacts fastball performance.
fastball<-pitching_joined %>%
group_by(Pitcher) %>%
filter(TaggedPitchType=="Fastball"| TaggedPitchType=="Sinker") %>%
select(avg_FB_velo,HorzBreak,SpinRate,InducedVertBreak, FB_wOBA) %>%
drop_na() %>%
summarize(avg_velo=mean(avg_FB_velo), avg_horz=mean(HorzBreak), avg_vert=mean(InducedVertBreak), avg_spin=mean(SpinRate), wOBA = mean(FB_wOBA)) %>%
arrange(-avg_velo)
fastball %>%
kbl(digits=3, col.names=c("Pitcher", "Avg Velocity (MPH)", "Avg Horizontal Break (IN)", "Avg Induced Vertical Break (IN)", "Average Spin Rate", "FB wOBA"), caption = "Fastball Data") %>%
kable_styling()| Pitcher | Avg Velocity (MPH) | Avg Horizontal Break (IN) | Avg Induced Vertical Break (IN) | Average Spin Rate | FB wOBA |
|---|---|---|---|---|---|
| Tross, Owen | 89.8 | 13.398 | 10.272 | 2182.094 | 0.318 |
| Dessart, Spencer | 89.3 | 6.903 | 15.573 | 2108.304 | 0.388 |
| Lessing, Justin | 87.2 | 10.015 | 17.767 | 1974.918 | 0.349 |
| Gatland, Andrew | 85.8 | 12.788 | 15.202 | 2042.858 | 0.384 |
| Duarte, Sawyer | 85.2 | 15.416 | 15.727 | 2221.194 | 0.458 |
| Price, Alex | 84.6 | -7.343 | 15.191 | 2049.501 | 0.383 |
| Mosier, Freddy | 84.5 | 16.074 | 8.513 | 2045.459 | 0.325 |
| Atwood, Jackson | 84.4 | 8.691 | 13.921 | 2136.420 | 0.306 |
| Knightly, Dylan | 84.2 | 20.498 | 6.749 | 2104.656 | 0.480 |
| Ritch, Alec | 83.9 | 16.037 | 7.259 | 1967.930 | 0.417 |
| Gustavson, Henry | 83.8 | 14.238 | 10.016 | 2146.486 | 0.213 |
| Handa, Kunal | 83.8 | 7.243 | 11.694 | 1801.582 | 0.240 |
| Crider, Cole | 82.8 | 12.621 | 11.967 | 1885.096 | 0.495 |
| Gattuso, Gavin | 77.9 | -14.476 | 11.686 | 1949.736 | 0.960 |
MiddPitch %>%
ggplot(aes(x=avg_FB_velo, y=FB_wOBA)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("Average Velocity")+
ylab("Fastball wOBA")+
ggtitle("Average Fastball Velocity (mph) vs. Fastball wOBA")
\[ \hat{FBwOBA} = 3.42 - (0.04)(avgFBvelocity) \]
fit3 <- lm(FB_wOBA ~ avg_FB_velo, data= MiddPitch)
tab_model(fit3) | FB w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 3.42 | 1.10 – 5.74 | 0.007 |
| avg FB velo | -0.04 | -0.06 – -0.01 | 0.015 |
| Observations | 16 | ||
| R2 / R2 adjusted | 0.356 / 0.310 | ||
Because there is a p-value of 0.02, average fastball velocity is statistically significant when predicting Fastball wOBA. The faster pitchers throw their fastballs, the better they will perform against Middlebury’s opponents. While the model is a strong fit, there is a group of pitchers that throw fastballs in the 83-84 mph range that see much greater success than predicted based on the regression model. This is explained by the deceptive qualities in their deliveries and pitch mixes. Kunal Handa has the second-best Fastball wOBA on the team but throws a fastball that averages just below 84 mph. While the number is not eye-popping, his fastball is perceived to be much faster to the hitter because he is 6’ 7,” so he releases the baseball much closer to the plate. Handa’s performance plays into what the regression analysis shows, which is how gains in velocity positively impact fastball performance.
MiddPitch %>%
ggplot(aes(x=FB_Spin_Rate, y=FB_wOBA)) +
geom_point() +
geom_smooth(method='lm', se=FALSE) +
theme_classic()+
xlab("Average SpinRate")+
ylab("Fastball wOBA") +
ggtitle("Average Fastball Spin Rate vs. Fastball wOBA")
fit4 <- lm(FB_wOBA ~ FB_Spin_Rate, data=MiddPitch)
tab_model(fit4)| FB w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 0.99 | -0.92 – 2.91 | 0.281 |
| FB Spin Rate | -0.00 | -0.00 – 0.00 | 0.518 |
| Observations | 14 | ||
| R2 / R2 adjusted | 0.036 / -0.045 | ||
Based on this regression analysis, average fastball spin rate does not influence fastball wOBA for the Middlebury pitching staff. I was surprised to see that spin rate does not influence fastball wOBA because spin rate is often associated with velocity (as velocity increases, spin rate tends to increase). I suspect this is seen for the Middlebury pitching staff because there are a significant amount of pitchers who do not spin the ball efficiently. Sawyer Duarte has the highest spin rate for his fastball on the team with a fastball that averages just below 86 mph, with an average spin rate that is considerably higher than both Owen Tross and Spencer Dessart who throw fastballs that routinely hit 90 mph and higher. Sawyer’s inconistency in performance, despite his high fastball spin rates, shows the importance of velocity and execution of his fastball in addition to other quality pitches in his arsenal.
Slider Stuff vs. Performance
The first step in my slider analysis was to break up sliders into two categories, sweepers and hard sliders. Because sweepers and hard sliders have very different properties as pitches, it is important to separate them for my analysis despite them both being tagged as sliders in Middlebury’s TrackMan data. Because sweepers and hard sliders’ primary difference is the amount of horizontal break, I decided, based on the data, that sweepers would have 7 inches of horizontal break (sweep) or more and anything below to be called hard sliders.
Slider <-pitching_joined %>%
group_by(Pitcher) %>%
filter(TaggedPitchType=="Slider") %>%
select(avg_SL_velo,HorzBreak, InducedVertBreak, SL_wOBA) %>%
drop_na() %>%
summarize(avg_velo=mean(avg_SL_velo), avg_horz=mean(HorzBreak), avg_vert=mean(InducedVertBreak), wOBA = mean(SL_wOBA)) sweeper <- MiddPitch%>%
select(Pitcher, avg_SL_velo, SL_Horz, SL_Vert, SL_wOBA) %>%
drop_na() %>%
filter(SL_Horz <= -7)
sweeper %>%
arrange(SL_wOBA) %>%
kbl(digits=3, col.names = c("Pitcher", "Avg Velo","Avg Horizontal Break" ,"Avg Induced Vertical Break", "SL wOBA"), caption= "Sweeper Data") %>%
kable_styling()| Pitcher | Avg Velo | Avg Horizontal Break | Avg Induced Vertical Break | SL wOBA |
|---|---|---|---|---|
| Handa, Kunal | 73.7 | -14.007 | 3.417 | 0.000 |
| Crider, Cole | 74.5 | -11.050 | -7.178 | 0.199 |
| Gatland, Andrew | 75.2 | -16.078 | 2.562 | 0.346 |
| Knightly, Dylan | 74.6 | -10.116 | -1.158 | 0.352 |
| Gustavson, Henry | 72.5 | -14.112 | -5.277 | 0.697 |
sweep <- lm(SL_wOBA ~ SL_Horz, data=sweeper)
tab_model(sweep)| SL w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 0.14 | -2.38 – 2.67 | 0.869 |
| SL Horz | -0.01 | -0.20 – 0.18 | 0.836 |
| Observations | 5 | ||
| R2 / R2 adjusted | 0.017 / -0.311 | ||
My first analysis of sweepers examined how horizontal break impacts their performance. Based on the regression analysis, the amount of sweep does not impact the performance of the pitch. I think this finding can be explained in multiple ways. First, there are likely diminishing returns with the amount of sweep. If a hitter recognizes a pitch with a ton of sweep (e.g., 15+ inches of sweep), there is a decent chance that the hitter will not swing because of how much it moves. Second, if the pitch has a significant amount of sweep, it may be challenging for the pitcher to throw the sweeper for strikes. Because home plate is 17 inches wide, a pitch with 15 inches of horizontal break may be difficult to land within the strike zone. The combination of hitters not swinging at the pitch and pitchers struggling to throw sweepers for strikes likely results in there not being a significant relationship between the amount of sweep and its wOBA.
sweepvelo <- lm(SL_wOBA ~ avg_SL_velo , data= sweeper)
tab_model(sweepvelo)| SL w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 8.50 | -21.31 – 38.31 | 0.431 |
| avg SL velo | -0.11 | -0.51 – 0.29 | 0.447 |
| Observations | 5 | ||
| R2 / R2 adjusted | 0.203 / -0.063 | ||
My second analysis of sweepers shows a lack of statistical significance between sweeper velocity and sweeper wOBA. The lack of a relationship makes sense because sweepers are inherently a much slower pitch due to their increased side spin. In future studies with a larger sample, I would be interested to see if there is statistical significance between sweeper performance and sweeper velocity because throwing pitches harder tends to help pitchers perform better.
hardslider <- MiddPitch%>%
select(Pitcher, avg_SL_velo, SL_Horz, SL_Vert, SL_wOBA) %>%
drop_na() %>%
filter((SL_Horz > -7 & SL_Horz <=7 ))
hardslider %>%
arrange(SL_wOBA) %>%
kbl(digits=3, col.names = c("Pitcher", "Avg Velo","Avg Horizontal Break" ,"Avg Induced Vertical Break", "SL wOBA"), caption= "Hard Slider Data") %>%
kable_styling()| Pitcher | Avg Velo | Avg Horizontal Break | Avg Induced Vertical Break | SL wOBA |
|---|---|---|---|---|
| Lessing, Justin | 80.1 | -0.463 | 0.369 | 0.242 |
| Dessart, Spencer | 79.6 | -2.573 | 3.766 | 0.246 |
| Tross, Owen | 76.6 | -5.955 | -0.890 | 0.248 |
| Price, Alex | 78.2 | 2.377 | -3.704 | 0.258 |
| Mosier, Freddy | 77.1 | -5.168 | -1.556 | 0.282 |
| Atwood, Jackson | 75.6 | -1.001 | 0.130 | 0.344 |
| Duarte, Sawyer | 78.2 | -4.043 | 5.607 | 0.357 |
slidevelo <- lm(SL_wOBA ~ avg_SL_velo , data= hardslider)
tab_model(slidevelo)| SL w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 1.45 | -0.89 – 3.79 | 0.173 |
| avg SL velo | -0.01 | -0.05 – 0.02 | 0.257 |
| Observations | 7 | ||
| R2 / R2 adjusted | 0.247 / 0.096 | ||
My first analysis of hard sliders examined how average slider velocity impacts its performance. Surprisingly, there is no relationship seen in my data between the two variables. I think this is caused by two factors. First, some pitchers throw sliders with pitch profiles closer to cutters due to the increased backspin on the pitch, resulting in increased levels of induced vertical break. Sawyer Duarte is an example of one of Middlebury’s pitchers who throws one of the faster sliders on the team but has the most induced vertical break on the pitch by far. Because his “slider” moves like a cutter (lack of vertical drop/depth in the pitch), it results in fewer swings and misses and soft contact. This causes Sawyer’s slider to be hit consistently harder and for a higher wOBA. I believe if Sawyer’s slider averaged an induced vertical break closer to 0 to 3.5 inches, it would perform much better.
slidevert <- lm(SL_wOBA ~ SL_Vert, data= hardslider)
tab_model(slidevert)| SL w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 0.28 | 0.23 – 0.33 | <0.001 |
| SL Vert | 0.01 | -0.01 – 0.02 | 0.338 |
| Observations | 7 | ||
| R2 / R2 adjusted | 0.183 / 0.020 | ||
My second analysis for hard sliders examined how the induced vertical break of the slider impacts the performance of the pitch. There is no significant relationship between the two variables. This finding makes sense because many hard sliders have gyro profiles, meaning they have an induced vertical break right around 0 inches. Because there are so many well-performing sliders right around the 0 inches of induced vertical break line, slight changes in the induced vertical break will not see similar changes in wOBA, resulting in the lack of a relationship between induced vertical break and slider wOBA seen in this regression analysis.
slidehorz <- lm(SL_wOBA ~ SL_Horz, data = hardslider)
tab_model(slidehorz)| SL w OBA | |||
|---|---|---|---|
| Predictors | Estimates | CI | p |
| (Intercept) | 0.28 | 0.21 – 0.35 | <0.001 |
| SL Horz | -0.00 | -0.02 – 0.02 | 0.799 |
| Observations | 7 | ||
| R2 / R2 adjusted | 0.014 / -0.183 | ||
My final analysis for hard sliders examined how horizontal break impacts the performance of the pitch. Based on the results, the amount of horizontal break does not have a significant relationship with hard slider wOBA. Because hard sliders tend to have gyroscopic spin, the horizontal movement of the pitch tends to fluctuate right around the 0 to 5 inches of sweep seen in Middlebury’s pitching staff. Due to these fluctuations, there is no relationship between the horizontal movement of the pitch and its performance.
Limitations
One of the main limitations of my research is the sample size of the data. Middlebury College only had 16 pitchers pitch this year, and many of the pitchers only threw in around 10 innings each, so singular performances can strongly impact a given player’s overall season statistics, which would heavily impact my results. However, I believe my findings would either hold up or my models would be better fits with a larger sample size of pitchers and more innings pitched for the players I examined.
Conclusions
My analyses show the importance of a pitcher’s overall stuff when predicting their performance on the mound. Fastball velocity and CSW% were statistically significant when predicting both FIP and wOBA, showing the importance of a pitcher having good stuff on the mound. I believe my analyses prove command is a means to maximize a pitcher’s stuff and should not be solely relied upon.
Based on this project, I would suggest that pitchers spend most of their time in the off-season training to increase the velocity in their pitches. Because fastball velocity was such a strong indicator of the success of our pitching staff, any growth in fastball velocity would have the largest impact on performance on the mound when compared to other adjustments that pitchers could make in the off-season.
For future research relating to this topic, I would love to have multiple years of pitching data at Middlebury College’s Division 3 level. I believe having more data would strengthen the findings found in this study.
Acknowledgements
I would like to thank Emily Malcolm-White (Lecturer in Mathematics) and Alex Harter (Assistant Baseball Coach/ Pitching Coach) for their help in my research.