Welcome to the Pitch Optimization Project
Late into a singularly heroic performance of a distinctively historic career, Pedro Martinez stood on the mound at Jacobs Field in Cleveland, staring down Manny Ramirez in the batter’s box. It was the bottom of the 7th in the deciding game of the 1999 American League Division Series, and Pedro Martinez had returned from an injury in Game 1 to pitch 3 2/3 innings of no-hit baseball in relief up to this point, but was facing one of the most gifted batters of his generation in Ramirez.
Martinez threw a high curveball, a curveball low & away, a fastball low & outside the strikezone, a changeup in on Ramirez’s hands, then another breaking ball low & away to bring the count to full at 3 balls, 2 strikes. At this point, Ramirez had seen a wide array of locations and pitch types, covering most of Martinez’s repertoire. For the 6th pitch of the at bat, he could try to change Ramirez’s eye level from the last pitch, or disrupt his timing with a fastball to counter his most recent breaking ball. Martinez’s signature pitch was his changeup, but he had just thrown an off-speed pitch with his curveball, so a changeup would be at the same tempo that Ramirez had just seen.
On the other side of the battle, Ramirez was battling to help his team stay in the game. If he could reach, he could bring the tying run closer to the plate and break-the-seal on Martinez’s shut-down night so far. With 2 outs and nobody on his priority was just to get on-base, which meant protecting the plate with a full count. Martinez could come at him with another breaking ball, but Martinez would know that’d run the risk of breaking out of the strikezone, and Ramirez had just seen such a pitch so had a good eye for it.
Martinez needed to throw a perfect pitch to escape a dangerous hitter and close the inning. Ramirez needed to anticipate and adjust to whatever Martinez was going to throw to give his team a chance to keep their season alive. Would Martinez throw another breaking ball away? Would it be a fastball up in? One of the hundreds of other possibilities? Only Martinez knew, and only Ramirez could react¹.
The Inherent Conflict
This is one example of the most foundational, but complex interplay in baseball: the pitcher² vs. hitter head game. This conflict occurs literally hundreds of times a night for hundreds of nights a year in Major League Baseball, yet we take the complexity and nuance of what is happening silently and invisibly for granted.
The consequences of this conflict could not be greater in baseball. There is literally no other action that has a greater effect on the outcome of a single at-bat, and thus, the game, than the result of the hitter’s reaction to the pitcher’s pitch. Anticipate AND react correctly, and the hitter get’s a 400 ft. homerun. Anticipate OR react poorly, and the swung bat misses the ball entirely, or the hitter watches a called strike pass by. The result of this conflict unilaterally determines what kind of batted ball (if any) goes into play, which determines how much value (or runs) the hitter created in the at-bat.
This interaction is as old as baseball, but unlike almost every other part of the game, our understanding of it has advanced very little. We can know the total value a player has created with each at bat, we can know which outfielders have the fastest reaction time and cover the most ground, but we have very little understanding of pitch selection helps or hurts a team win.
On its face, the conflict is simple: the pitcher is constantly trying to confuse the hitter, and the hitter is trying to divine the inner machinations of the pitcher so he can anticipate and react to the pitch coming his way. Assuming that most major league hitters have the physical ability to hit most pitches from major league pitchers if given perfect information, confusion becomes the pitcher’s only real weapon. The pitcher who can confuse the hitter the most and for the longest period of time will have the most success.
However, achieving max confusion is easier said than done. Just as each pitcher has relative strengths, so do hitters. How these strengths and weaknesses interact creates a highly complicated collision of abilities with every at-bat. Additionally, we can easily fall into the “I know that he knows that I know that he knows I’ll throw a fastball, so I should throw a curveball” game theory doom loop that befalls games of anticipation and reaction.
So how do we endeavor to better understand this vortext of choice, physical ability, and reaction? How will we improve our ability to answer the question “What is the best pitch to throw right now?” With data. Lots and lots of data. Welcome to the Pitch Optimization Project.
Any given baseball game has about 300 to 400 pitches thrown between both sides. Over the 2,430 Major League Baseball games played per year, that is almost a million pitches to analyze per season. And with the advent of statcast, we now have several seasons of exhaustive data on every one of those million-pitches-per-season. Our hypothesis is that trends will emerge in that data that will reveal insights into how pitch sequencing can affec the outcomes of at-bats and games. Armed with this understanding, we can replace mystery and guessing with data and science.
1 With the count full, Pedro Martinez threw another fastball up-and-in to Manny Ramirez, which he caught up to, barely, and fouled off. The next pitch was another fastball, but this time in the opposite corner of the strikezone, low-and-away, which froze Ramirez for a called third strike. Pedro Martinez went on to complete 6 innings of no hit baseball and help the Red Sox advance to the ALCS.
2 We know pitchers actually have very little involvement in choosing which pitch to throw, but for the sake of simplicity, we’ll group the coaches, catchers, and pitchers into a single group that we’ll refer to as ‘the pitcher’