Pitching Backward: First-Pitching to the Situation

A few weeks ago I wrote about James Shields’ success with runners in scoring position, specifically noting how it coincided (causation or not) to a change in the righty’s approach. This led down a deep, dark rabbit hole in which the BP research staff helped me compile a more exhaustive list of how pitchers changed their tendencies in these situations. The full data set includes the absolute changes in pitch selection in several different count situations (first pitch, pitcher ahead, count even, batter ahead, etc.); for this particular post, though, I’ll focus on just one of these subcategories: first pitches.

I should disclose before we get too far into this that I’m honestly not sure if the data we’ve mined is truly descriptive and valuable. In theory it is both descriptive and valuable, but it’s very possible that there’s a better way to interpret and process this data. Or, perhaps, that there’s a wrong way to interpret and process this data, and that I’m doing it. So with that, I’ll explain the methodology for this analysis.

The BP research team scoured our PITCHf/x database, pulling the pitch-type tallies in each ball/strike state for every major-league pitcher. From there, the pitch counts were broken down into two base situations: runners in scoring position and no runners in scoring position. With these pitch counts we were able to break out pitch usage percentages and compare them to one another. Adding the absolute values of each change, we gave the pitchers all “change in approach” scores. For example:

Pitcher A throws two pitches: a fastball and a slider. On the first pitch of an at-bat with no runners in scoring position, he throws 65 percent fastballs, 35 percent sliders. But when a runner reaches second and/or third, 75 percent of Pitcher A’s first-pitches are fastballs, and 25 percent are sliders. The 10 percentage point increase, and the 10 percentage point decrease, add up to a 20-point change. We repeated this process for different ball/strike situations—pitcher ahead, count even, batter ahead—and added those changes. This gave every pitcher one total change score, and various sub scores. (The full data set—more than 600 pitchers—was trimmed down at the cutoff of 250 pitches in each scenario, with the goal of removing most non-regular relievers. Different pitch count cutoffs would give you a slightly different leaderboard, but the end result for relevant pitchers was about the same.)

To move from the hypothetical example to real examples, we can illustrate the process with two pitchers at opposite ends of the spectrum. The pitcher who most changed his pitch usage when he got into trouble was Hisashi Iwakuma, while Josh Collmenter was last among starting pitchers. Iwakuma’s first-pitch change score when runners reached scoring position was about 83 percentage points. Collmenter’s pitched selection in the same situations was around 5 percentage points.

To see what that looks like: Below are pitch usage charts for each pitcher—no RISP on the left, RISP on the right. These side-by-side images make comparisons between the two base situations easy–well, easier. Click to enlarge for true ease:

Collmenter:

Iwakuma:

For Collmenter, the biggest change in first pitch approach with RISP is a 3 percentage-point hike in changeups. He throws slightly fewer cutters, and an almost imperceptibly smaller number of curves

Iwakuma, meanwhile, cuts down on his first-pitch sinker use by 30 percentage points, and his four-seamer by about 8 percentage points. Most of those pitches become splitters or sliders.

Those shaded boxes in the Brooks Baseball charts signify more extreme adjustments to the situation: Red means an increase of at least 10 percentage points above his baseline pitch usage, blue for a decrease of the same size. Collmenter’s chart is quiet. Iwakuma’s could be a Mondrian painting.

It’s worth noting that pitching one way isn’t better than the other. Sure, the excellent Iwakuma and his ace teammate Felix Hernandez are the two pitchers whose first-pitch approach changed the most when runners reached scoring position. But Clayton Kershaw made the fifth-smallest adjustment, according to this method of measuring. For greater context, here are the pitchers whose first-pitch approaches changed the most depending on the base situation*:

The AL West shows up quite a bit in this list with Iwakuma, Hernandez, Gray, Wilson, and Ramirez all representing. Meanwhile, the pitchers who had the smallest changes in pitch:

It is worth noting that this list is dominated by relievers, but Ryu, Hughes, and Kershaw were all good or great with an “unnuanced” approaches, reinforcing the reminder that changing an approach isn’t necessarily a good or a bad thing—it’s just a thing. For what it’s worth, the pitcher I profiled a few weeks ago came in at 98^th on the list. While he’s still in the top half (238 pitchers cleared our minimums), Shields ended up a lot lower on this list than I expected based on the analysis I had done previously.

This point is important though, because while the changes in Shields’ approach for the two base situations (a 26.1 percent change on average) seemed significant, there were 97 other pitchers who changed more significantly. This sheds light on how little research has been done on pitch usage changes, a worthwhile topic for understanding why some pitchers might perform better or worse in a wide variety of situations.

Let’s take a look at how this comes to life for Iwakuma. All of the following GIFs come from the bottom of the second in Iwakuma’s September 21^st start against Houston, beginning with Jake Marisnick at the plate with one out:

Iwakuma starts Marisnick off with a sinker, his more typical first pitch approach. With no runners in scoring position, half of his first pitches to righties are sinkers. Marisnick took it for a strike. Later in the at-bat, Marisnick would line a single to left. Next up was Gregorio Petit, a right-hander:

Petit would also get a sinker—51 percent of the time, a right-hander facing Iwakuma would. He hit a hard grounder for a single. Marisnick moves into scoring position, and our expectations for Iwakuma’s first pitch change dramatically. Up steps switch-hitter Jonathan Villar:

As Petit did, Villar attacks the first pitch—but this time it’s a splitter. With nobody on, Iwakuma throws first-pitch splitters to lefties just 13 percent of the time. With RISP, it spikes to 50 percent. Maybe the advance scouting reported noted this and it didn’t help Villar, or maybe it didn’t note it at all: Either way, he grounds out to second. Two down; runners on second and third. Next up for the Astros is leadoff hitter Robbie Grossman.

Grossman takes the offering for a ball as Iwakuma buries a splitter in the dirt. Once again we see Iwakuma going with a split-fingered fastball on the first pitch of an at-bat. Five pitches later he strikes out Grossman with another splitter in the dirt, ending the inning.

This inning is more of an in-a-nutshell GIFbomb than anything we can draw conclusions from. Even if we could draw conclusions—like, look at how Iwakuma generates early weak contact with the splitter, or something along those lines—they would just raise more questions. For example, why wouldn’t Iwakuma throw more first-pitch splitters all the time if it’s so much more effective than going sinker first? And, In fact, Iwakuma was about 55 percent worse with runners in scoring position than overall this season. Was tweaking his approach so significantly an optimal strategy at all?

So no, no conclusions. More than anything this analysis raises further questions about why pitchers alter their repertoires, and what impact such volatility might have on performance. For Iwakuma, changing approach as runners advance around the bases is standard operating procedure. We can’t say whether the change is good or bad. We can say, at least, that it’s there, and that’s a start.

*Note: There are small differences between the numbers in the charts and the numbers in our math, as those in the data set we pulled do not exclude intentional balls and pitchouts.

Tremendous thanks to Rob McQuown for statistical research, and to Sam Miller and R.J. Anderson for their counsel.