The Junk Box Is Full of Mystery

Every once in a while, I take a peek at the Kirby Index leaderboards. In August, the expected names that populated the top of the rankings: Kevin Gausman, Trevor Rogers, and Jacob deGrom. Similarly, the bottom featured stereotypically wild hurlers: Joe Boyle, Luis Gil, Freddy Peralta, and… Janson Junk???

If you’re familiar with Junk, it’s most likely due to his excellent command: Among all starters with at least 70 innings pitched this season, his 3.0% walk rate ranks as the lowest. (In an excellent interview with David Laurila last month, Junk talked about training his command at Driveline over the winter.) Seeing Junk in last place on this leaderboard was like spotting a polar bear in Arlington, Texas — in other words, a sign that something was seriously amiss.

By all four components of the Kirby Index, Junk ranked poorly. But his vertical release point was particularly inconsistent, sitting dead last among pitchers in the sample. In my article from last year introducing the Kirby Index, which I linked to above, I found that the ultimate location of the pitch is dictated almost entirely by release angles and release points. It follows that pitchers with inconsistent release points exhibit poor command. How was Junk varying his release point so frequently and still throwing so many strikes?

Paul Skenes is not a similar pitcher, but he provided a potential answer. Skenes showed up near the bottom of original Kirby Index leaderboards because he changed his position on the rubber depending on the handedness of the batter. Ultimately, that was an easy fix: By looking at fastballs thrown to right-handed hitters and left-handed hitters separately and averaging the results, the rubber shifters could be compared to the rest of the pitching population.

However, Junk wasn’t shifting around the rubber, and even if he was, it wouldn’t change his vertical release point. But adjusting his arm angle would. Arm angle wasn’t publicly available when I first wrote about the Kirby Index, but it turns out that for Junk, it explains the significant variation in his vertical release point as well as his vertical release angles.

From one pitch to the next, Junk can switch from a milquetoast three-quarters release…

… to one of the more extreme over-the-top arm angles in the league, landing somewhere in the Jonah Tong/Trey Yesavage zone.

Here’s the three-quarter still:

And the over-the-top one:

Among starters with at least 300 fastballs thrown, only Antonio Senzatela and Logan Allen featured more arm angle variation on their four-seam fastballs.

While I didn’t have access to arm angle variation when I wrote that Kirby Index article, this funky Junk fact motivated me to see if this related at all to the ability to command the ball. In short: Apparently, not at all!

To test this, I employed our two trusty models, Location+ and PitchingBot’s BotCMD, as proxies for command skill. I used the standard deviation of a pitcher’s four-seam fastball arm angle as the explanatory variable; for each respective model, I used both the overall grade (Location+ and botCMD) as well as their fastball-specific models (botCMD FA, Loc+ FA) as the response variables. I limited the sample to pitchers with at least 60 innings pitched and 300 fastballs thrown. Across all four combinations of these explanatory and response variables, there was hardly a relationship.

OK, maybe if you squint, there’s something going on there with PitchingBot’s fastball command model. But on the whole, there’s really nothing there.

If arm angle variation doesn’t explain command, it must explain something, right? After all, there is a strong and proven relationship between arm angle and induced vertical break:

Justin Willard, the Red Sox director of pitching, delivered a relevant quote in The Athletic’s long article on induced vertical break published last week.

“Arm slot is the highest correlated thing to spin direction,” Willard said. “Spin direction is the highest correlated thing to Magnus lift.”

But something else is happening here that’s sort of surprising. When Junk’s four-seam fastballs are plotted against their induced vertical break, it becomes clear that for him, specifically, the relationship vanishes. Whether Junk throws from way over-the-top or from a standard three-quarters position, the amount of vertical movement he gets on his fastball hardly varies.

It turns out this phenomenon — while not as true for others as Junk — holds true across cases. While the r-squared between arm angle and IVB at the pitch level is 0.34, the averaged r-squared within pitchers themselves is 0.05.

This felt like another mystery in need of an explanation. How can the relationship on the broader level between IVB and arm angle be so strong, yet explain almost nothing at the pitcher level? I brought this question to my pitching pal Stephen Sutton-Brown at Baseball Prospectus, who provided a compelling explanation. At the individual pitcher level, arm angle and velocity remain (relatively) constant, and so the variation on any given fastball is explained more by other factors, like hand positioning or grip. At the pitcher population level, the arm angle variation is, of course, much wider.

But Junk is a special case. He does feature a decent amount of arm angle variation, and so I expected some portion of his fastball movement to be explained by the changes in his arm position. It was not. Junk was one of the few pitchers with zero relationship at all between his arm angle and fastball vertical movement.

What’s going with this guy? I wanted to ask him. Sadly, the Marlins did not have the same desire; they declined to make him available for an interview. For now, the mysteries of Junk will remain out of reach.