Measuring Pitch-Arounds by Ben Clemens May 31, 2022 © Brad Mills-USA TODAY Sports On Sunday afternoon, Juan Soto stepped up to the plate in the top of the first inning with a runner on first base. Soto, as he is wont to do, took the first pitch. He took the second pitch, too, as Kyle Freeland struggled with his command. Freeland relented and threw a slider over the heart of the plate, middle-away, hoping to sneak back into the count. Soto hit it 400 feet for a home run, putting the Nationals up 2-0. When Soto batted to lead off the bottom of the fifth inning, Freeland was still pitching. Again, Soto got ahead 2-0. This time, Freeland was far more careful. He clipped the top of the zone with a fastball for a called strike one, then attempted to paint the corner low and away on his next pitch. He missed, and down 3-1, he threw another pitch low for ball four. Soto took his base, but the Nats couldn’t drive him home. Why did Freeland challenge Soto in the first? Why did he change his approach in the fifth? I can’t read minds, but the decision seems fairly straightforward to me. In the first, Freeland didn’t have the luxury of pitching around Soto; a walk would put a runner in scoring position. In the fifth, the situation wasn’t quite so bad; a walk put a runner on base, which isn’t ideal, but there’s something primally scary about walking a runner to second. That’s the theory, at least. It’s how I’ve understood baseball as long as I’ve watched it. Good hitter, base open, advantageous count? That hitter might as well send his bat back to the dugout, because he’ll rarely get a pitch to hit. Put that runner on first base, and the equation changes completely – now a walk hurts too much, and pitchers will take their chances in the strike zone. Rather than accept that conclusion as received wisdom, I decided to test it. But how? I couldn’t exactly watch every plate appearance and mark down “pitched around” or “not pitched around” – there’s no exact definition for what pitching around means, and even if my subjective judgment were good enough, I’m not going to watch that many plate appearances. I decided to create a strict test. First, I created my samples. In one corner, we have 2-0, 2-1, 3-0, and 3-1 pitches with a runner in scoring position but first base open thrown in 2021. In the other corner, we have those same counts, but with a runner on first base. Why add the condition that a runner has to be in scoring position, unlike the Soto example? I wanted to find obvious pitch-around situations. With a lesser hitter than Juan Soto, you might treat the bases empty and no one out as a time to attack, even against a good hitter. A runner on first base, after all, is a pretty bad result for the defense. From there, I measured the distance between each pitch and the exact center of the strike zone. Why not use Baseball Savant’s attack zones? The reason is kind of disappointing, but it’s a reason nonetheless: you can’t download which zone each pitch was in, merely search for them by zone, and I wanted to work with a set of all the pitches thrown in those counts rather than break them out by subset. As you’ll see, it also allows me to work with the data flexibly later on. With the two groups set up, I did a first pass by looking at the percentage of pitches that were within 8.4 inches of the dead center of the zone. Why 8.4 inches? That’s roughly pitches that are over the heart of the plate – the proportion of pitches within 8.4 inches of the center of the zone closely matches the proportion of pitches in the “heart” attack zone. It’s rounded instead of square, but if you’re looking for pitches down the middle, it’s a close fit. How do pitchers behave? With a runner on second and first base open, they threw 30% of their pitches “down the middle.” With a runner on first, they threw 31.5% of their pitches down the middle. Hey, look! Evidence that pitchers behave differently with a runner on first. But we can do better. Pitchers don’t treat every batter equally. I took Steamer projections for the 2021 season and split them into two groups: hitters with a projected wOBA above league average and hitters with a projected wOBA below league average. It’s not a perfect cut, but it roughly gets the idea I’m going for: separating good hitters from bad hitters. How did this extra filter do at detecting pitch-arounds? Incredibly well: Middle-Middle%, Hitter’s Counts Bad Hitter Good Hitter 1B Open 31.7% 27.8% 1B Closed 32.4% 30.6% Now we’re talking! Pitchers attack weaker hitters at more or less the same rate (read: with great aggression) when they get behind in the count whether there’s a runner in scoring position or on first base. That makes perfect sense: you don’t want to let bad hitters get on base cheaply. They’re below average for a reason. On the flip side, “don’t let a good hitter beat you in a hitter’s count” seem like words to live by, but not when there’s already a runner on first base. Another reason to measure by proximity to the center of the zone: you can vary the distance from center you’re measuring. Worried that my arbitrary 8.4 inches was cherry picking a cutoff that would make my data look good? Here’s the “first base closed aggression addition” – how many more pitches are dead red with first base occupied – for a variety of definitions of dead red. Of note, these are all in percentage points: First Base Closed Aggression Addition Inches Bad Hitters Good Hitters 6 0.3% 1.5% 6.5 0.5% 1.7% 7 0.2% 2.3% 7.5 0.6% 2.8% 8 0.4% 3.0% 8.5 0.9% 2.5% 9 1.1% 2.0% Why does the data start to blend at higher numbers? That’s reaching the fringes of the zone, particularly horizontally, which is starting to measure less whether a pitcher is nibbling and attacking and more whether they’re throwing strikes. That’s my story, at least, but if you have a different interpretation, that’s fine too – it’s pretty speculative data, and I’m doing a fair bit of twisting to get my results. It could just be random variation, or something else I haven’t thought of. Regardless, the general pattern holds: pitchers are measurably more careful against good hitters. One last twist: what if we only looked at elite hitters? I upped the cutoff from everyone with a higher than average projected wOBA to everyone with a wOBA projected to be at least 25 points above average, and the pattern persisted. In fact, the gap was even greater for “elite” hitters: Middle-Middle%, Hitter’s Counts Non-Elite Hitter Elite Hitter 1B Open 30.3% 27.1% 1B Closed 31.7% 30.8% Regardless of whether this is actionable, I think it’s kind of neat. I like when the numbers validate the eye test, and particularly when they can prove a small effect like this that isn’t easy to see in top-level rate statistics. If you think your favorite hitter is getting pitched around, you’re probably right. Just get him that classic lineup protection – by which I mean having a runner on first base – and he’ll be good to go. Oh, and just one last pedantic note to leave you on: the data on these won’t be perfect because of the way Statcast calculates the top and bottom of the strike zone, though I’m okay with that. On pitches where the batter didn’t swing, Statcast looks at the actual stance of the hitter and calculates a top and bottom of the zone. On pitches where they swung, it uses the average top and bottom of the zone. That might make for some messy data – but given that I’m just looking at this for fun rather than to tell you some player has demonstrated a new skill, I’m perfectly fine with that.