One Fastball Isn't Enough

David Butler II-USA TODAY Sports.
Every season seems to feature articles detailing the decline of fastball usage across professional baseball leagues; 2024 was no different. One team that stood out during this conversation were the Red Sox who seem to employ an intriguing pitching philosophy centered around Spin go brrrr. Though their four-seam fastball usage trails by quite some margin, their sinker usage ranks ninth overall and cutter usage ranks first overall; all this makes me immensely curious! For proof, just check out this graph!

Early counts (0-0, 0-1, and 1-0) where batters tend to chase fastballs have seen four-seam fastball usage drop significantly; that matches up perfectly. Conversely, sinkers and cutters have returned to levels last seen in 2018. So really the Red Sox don’t represent an isolated trend but are simply reflecting what many teams in MLB believe – that the increase exists even without Boston as part of that storyline.
Why is that? A look at metrics is less clear: Early count four-seam fastballs tend to return higher xwOBA values than sinkers and cutters – something which has always been true since these pitches first entered play. Sinkers and cutters have not performed better relative to previous iterations of themselves; an interesting theory holds that for pitchers with average or below-average velocity, sinkers tend to be an advantageous pitch choice. Driveline Baseball’s updated Stuff+ model revealed that sinkers averaged higher stuff ratings than four-seammers up until 97mph; thereafter, four-seamers took over exponentially. As healthy pitchers who reside within this velocity band remain somewhat rare, more conventional “normal” pitchers could turn toward sinkers to reduce hard contact risks.
Today I wanted to share my answer. Although initially my objective was analyzing pitch sequencing strategies and their effect, baseball often provides surprises when we least expect it. Sequencing appears intrinsically linked with fastball usage.
At first, I collected pitch-level data from both 2022 and 2023 seasons. Through careful study of every plate appearance’s pitch sequence – from first pitch popups to 15-pitch stalemates and beyond – this provided my data set with everything it needed. As part of my project, I made several major (yet contentious) decisions. First and foremost was to exclude plate appearances which resulted in hit-by-pitches (made possible due to poor control rather than sequencing issues) hit-by-pitches were an easy choice since these errors usually involve poor control rather than sequencing problems. Walks involve more of the sequencing aspect, since an incorrect order could force batters into swinging at balls off their plate. Still, I believe walks to be much more the result of command and stuff; after taking into consideration both hit-by-pitches and walks as possible outcomes, however, it would be more interesting focusing on balls in play as potential outcomes.
Second, instead of basing my calculations solely on literal outcomes, I calculated an expected run value for every batted ball based on factors such as its angle of attack, count, and whether or not its pitcher held platoon advantage. Doing this allowed for multiple factors: As pitchers possess greater control of launch angle than exit velocity, it makes sense to reward (or penalize) them accordingly. Also consider that inducing contact in unfavorable counts usually has greater run value on average than striking out. Platoon advantage extends beyond balls put into play. Righty-on-righty sinkers tend to frequent Salty Spitoon while righty-on-lefty sinkers may end up crying shameful tears at Weenie Hut Jr. One last change: I included sequence length as an independent variable which accounts for its moderate negative linear relationship to run value.
With my data in order, my model of choice was a random forest; this methodology uses multiple decision trees in conjunction with statistical techniques in order to make predictions. You might recognize a decision tree from its similarities to how pitchers make decisions regarding pitch sequence: Starting off with one possible pitch option and gradually expanding upon them as more possibilities emerge, much like when selecting pitch combinations in baseball. (If I were a professional pitcher, my indecisiveness would likely be my downfall). To emphasize, though: machine learning is certainly not required here and may even be overkill, though I found building models very enjoyable (unlike querying data!). Much of my methodology for building models came directly from Dylan Drummey’s article here on how he approaches data visualization – thank you Dylan!
Now it’s time for fun. Say, for instance, your goal was to reduce damage upon contact; which first pitch should you select as part of an attempt at mitigating impact damage on contact? Here’s what the data tells us by way of model-predicted run values:

Starter Pitch 1 of 184 = Sinker (0.0184), Cutter 0.0300 and Four Seamed 0.0354 are Sinkers; Cutters = Cutter [0.0300], Four Seam = 0.0354 with Cutters/Cuters as Cutters (0.00300/0.0300 and Cutter=0.0500 for Four-seam pitches respectively]. Changeup=0.389 and Slider = 0.0037 in Curveball pitch count order respectively
Sinkers and cutters seem to be the way to go in our simulation, while four-seam fastballs, changeups, sliders, and curveballs tend to get hit harder by batters if any contact was managed at all. When querying actual sequences rather than modeling outputs this order also shows up; but perhaps you already knew this? While validating that our model wasn’t stuck somewhere obscure; let’s move onto more interesting two pitch sequences such as these instead:

Top Ten Two-Pitch Sequences

Pitch 1 and Pitch 2 of our RV Model are as follows. Initially we determined RV values as follows for Pitch 1, RV values were determined for Four Seam (0.0076), Sinker for Four-Seams of 0.0282 for Cutter #0.0196 with changes made between Cutter #2 (0.0266), Slider #0.0205 and Change Up at Cutter 0.0264 with four seam length of 0.0298 in Four Seam (0.0591 for Four-Seam), Sinkers with four seam seam width changes up to 0.0311 for Cutter
Starting off a sequence hasn’t changed much over the years; however, second pitches subvert expectations in their delivery and timing. Instead of breaking eye levels with offspeed or breaking pitches, instead fire off another fastball of different type to alter eye levels subtly instead of changing eye levels drastically – an attempt at subtle change that involves just two miles per hour and inches; somewhere between an old school coach who demands fastball pounding into zone and Gen-Z analysts seeking 10 breaking balls at once lies our solution – fastballs remain good but now have some subtlety that makes three pitch sequences similar results?

Here is our Top Ten Three-Pitch Sequences (TSSCs).

Pitch 1 Pitch 2 Pitch 3 and Predicted Mean RV. for both pitches (pitch 1, pitch 2 and pitch 3, respectively). Cutter, Four-Seam Slider and Curveball each received four consecutive strikes of four seam sinkers for pitches 1 through 3, each clockwise around the circle until pitch 3, when Curveball could complete his swing (0.0053 for five-seamer and five for slider), so four seam slides began at around five milliseconds per side in Cutter (0.0038, Cutter Cutter for four seam Slider in Four Seam [0.0085 | Four Seam to Sinkers 8.0085 | fourseam Cutter, 4 Seam Cutter Curveball (0.0156 Changeup and Five Seam Swelled to Sinker [50.085 | Four Seam; Slipper 8 5075 | Four Seam | 4 SeamSinker 7 Seim and Tension 616 (changeup =0.0159 11.521 4 Seam Cutter Curveball (0.0053,54”sinker in Cutter Cutter Curveball’s=0.0053] but also four seam Sinkers at fiveseam for four-Seamm, Four Seamm 574 [Wayne Wright) at least) umplut the following four Seim with four Seame =0.0156” Curveball”, Four Seam, 48th Wrear =50 572]
Eight 5 Slam ‘Th’sem=0.0085=0.0085/8500/1650 Wsam Whilst Four Seam, 5 7Atm =5055] The Sinkers ‘On 4SemWill with Four Seam =0.0057″. In reality 0.0153 Whilst 4Sem, whils 0.0057 per seam, Four Seamed; six seam 8 5 7T’ for four-Sem (CF 0.0072 Squ =TeM =FiveSeman = S.Neam = 5056 Ws =0.0085=Sem.85 (0.0109 (0.0085 = FourSem as 4Seamer=5Seamer (Nike’); FourSem Cutter CurveBALL), FourSem=0.0005 781 Washout 858=0.0057 ‘Lander’t) and somm (CUT 51 ‘CuTer85……………….57 =.0057=5 6th 85 =0.091Wamer =0.008589 8585=0.085 WPage- 1s) 54.0085=0.0085 575) Whilst =0.0085).789) 875 WPamer =0.0587) 573 85.5 7Wamer 75 for ‘ 6755Wamer =0.0585 15191-Sem ChangeUp 5 677), Eight Seman 5785+608 5 7575 573 Went FourSeme; (0.0085 61985=0.057)S) ‘
Unfortunately, here is where the model shows its shortcomings. Here’s my theory on what’s happening: Because sinkers are clearly superior pitches for contact suppression purposes, I suspect the model overcorrects and assumes any sequence that starts off with one is perfect; thus weakening correlation between predicted run values and actual ones with increased sequencing length. Since no solution could be found to address this, going forward we will also rely on empirical evidence:

IRL Top Ten Three-Pitch Sequences

Pitch 1 = Pitch 2 = Pitch 3 with respect to Actual Mean RV in terms of Sinkers = Cutter and Four-Seam Cutters= 5mm Sinker-cutter is at-0.0562 Cutter-4 Seam Sinkers at 0.0562 which correspond to Four Seam Sinker Cutters being set between 0.0482 to 0.0562 for changeup, Slider to Curveball-0.026266 Cutter-3 Seam Cutter with changeup at-0.026266 = Curveball Changeup= 0.0266 Cutter- Curveball Slicker Changeup Cutter: Shaft Cutter with changeup Curveball-0.0098 Veranderung: Sinker for Four Seame Curveball-0.0098 Cut Up Cutter, Four Seamed Sinkers to Cutter and Cut up at-0.0566 for Changeup Cutter; Four Seamed Curveball by Changeup Cutter-4 Seame Cuveballot Changeup with changeup Curveball=0.026266 = Changeup Curveball0.049266 Cutter to Curveball changes around which changes could make up Curveball around this number that’s had Changeup Curveball to Curveball at0.026266 Change Uprov TO Curve Ball [0.026266 =0.049266 =0.049266 =0.026266 Slider =0.049266 Cuttler, Four Seamed to Changeup Curve Ball=0.0188 3 Change up and four seam Cutter Slider, Change up Curve Ball=0.026266 Slder=0.0193]. Change Up CurveB 18 23 193 Change up Culer and Change up CurveB=0.0098 but ‘slbe 0.0 193 [Slver =0.0193 19 3 193. “Changeup=0.0088 =0.0195 3=0.00 -193]. Sinkers 4 Seamed CurveB =0.000768”, when Slame 1885]. 6A=0.076 266 Cutter- 1023] Cutter). CurveBat 788 “.0066 Change-Slver=0.00066 for 4Samed 3″, Cutter CurveBALL and ‘Change up CurveB”Cr to CurveB 1894=0.126266 ‘ Change ups +0.0126) Cutter four Seamed in 2016 whil 0.0 262. 26Curr=0.000 983 ‘Cu 3.0094 +0.0 0.0 0.0262 CUTter ‘Change”, 471=0.0596 and this Trough 18-193 [ Cutter | Curve=0.0096 3 193, Four-0.01468 =0.009588″. 973). 193, Four seam, 892 and ‘change 0.0196” ChangeUp/0.01903 =0.1 262…….418 3 (0.056] 872 whils 133/0.0 193, 18 ‘Cuuter (0.0098 1”=0.193=0.0098 Cutter 8Sander 468 for 1889 1 262/0.0126 1/1873=0.00066]. Change Up 1895/0.048266/ 0.156 for 193/935). ChangeUp and 2143 for 16193-193 193=0.009888 487…..16).1893 but Change “Cor 668266 and 193…..2098 CUT 489/193C). 4 76% but 18 193/ 193 to 193 193/0.0190), but 1”. 318 193/0.026698″ in 13 CUT 193, CurveB3 193, Change to 193 200.0 193.49
These results seem more realistic, with more sliders, curveballs and changeups appearing, reflecting how an average pitcher might navigate an at-bat. And yet…

Eight out of 10 initial pitches consist of either a sinker or cutter; only two non-fastball pitches were encountered – there wasn’t even one four-seamer at play!
Instead, four-seamers can often be seen roaming second pitch land; this indicates once more the importance of switching up fastball types; six of 10 second pitches were fastballs in this regard.
Third pitches show an unexpected twist: now breaking and offspeed pitches dominate, while three fastballs might be considered excessive; sinker-cutter-sinker is currently the number one sequence. Overall, fastball is prominent.

But let’s think for a minute – if these observations are indeed accurate, why is there such an extensive attempt at eliminating fastball pitches from baseball’s repertoire?
Drummey reached an opposite conclusion from my analysis; that curveballs and sliders had lower predicted run values while cutters/sinkers did. Of course this doesn’t indicate either one of us has made catastrophic mistakes! At the root of it all lies pitching’s curse of contact suppression (measured by launch angle) is often bad at creating swings and misses; for instance, sinkers lead all other pitch types in terms of groundball rate year after year. At the same time, they rank last in terms of whiff rate and chase rate and as recent bat tracking data has demonstrated, it’s easy for batters to square up. As I did, excluding strikeouts, walks and exit velocity would result in overrating sinker pitches while underrating those like sliders that offer something the sinkers don’t; leading all pitch types in both whiff and chase rate and making batters struggle to square up for contact. Via Alex Chamberlain:

Sinkers don’t always reign supreme and therefore dominate league play; sliders are far superior from an inherently quality standpoint to sinkers; thus an ideal pitcher would throw only sliders to hitters on the outer edges of the plate in every outing and bask in glory every outing ideally; unfortunately life doesn’t work like that: some pitches touch corners while others land squarely in the middle; breaking balls need fastball setup before reaching their full potential and not all batters respond as intended to spin; eventually you may need to throw a fastball and hope it makes an impression; thus it pays to make smart choices about what fastballs, how often, when, where and when throwing fastballs might.
Certain aspects of “how” have already been tackled – for the most part, league consensus exists on both high fastballs and throwing fewer of them overall – yet many details regarding what and when remain unclear. Our model’s results provide some illumination; specifically that pitchers looking to minimize contact – in terms of potentially detrimental launch angles – by adopting multiple fastball types such as four-seamers, cutters and sinkers as a means of accomplishing this aim are best off embracing an assortment of fastball types from across this range in their repertoires.
Order is important; fastball triad should be employed early in a count rather than when batters fall behind. Looking at combined sinker/cutter usage during counts that favor pitcher, we see that combined sinker and cutter usage during these counts has reached its lowest ever point compared to early count fastball rate; this gap demonstrates some teams are employing sinkers/cutters first followed by changeups/sliders when attacking.
Remarkably, pitching backwards echoes more traditional approaches. Pitchers once established counts with fastballs before switching over to strikeout pitches once two strikes had been earned – yet today this reverse approach to pitching has only recently gained prominence despite already existing in past days – although typically done through four-seam fastballs rather than sinkers or cutters; hence making its execution part of modern game strategy rather than traditional tactics.
Reconsidering my original question, the league doesn’t appear to be trying to banish the fastball; rather, there appears to have been an attempt by its teams and clubs to redefine its definition. When looking at it as only an effective primary pitch and only using that fastball for two strikes against someone it becomes natural that pitchers would rely solely on that pitch type; but by viewing fastball as more versatile construct, pitchers might start thinking more creatively and using different kinds of fastballs together may prove more successful than using just one type alone.
Since pitch usage depends upon which players are on which roster at any one time, I am reluctant to make broad statements regarding teams or leagues in general; however, one concept I strongly support is having multiple fastballs at our disposal; although this won’t significantly change anyone’s life or paradigm shift them radically; having additional fastballs could actually aid pitchers who typically struggle keeping the ball down – evidence exists showing its effects – without needing extensive mechanical upgrades; instead it just needs divert opposing hitters’ attention away; Tyler Glasnow’s experimentation with sinkers is great example!
Fastballs have returned as part of the kitchen sink’s arsenal. Although traditional fastball usage may no longer exist, fastball pitches continue to appear with new and innovative outfits on them.