23 August 2017

How Pitch Sequencing Can Affect Outcomes

The 2017 Baltimore Orioles starting rotation has been a bit of a mess. Maybe with some data and some advance planning, the starting pitchers can become more effective. If we can identify the two-pitch sequences to end plate attempts and compare outcomes to those sequences, we may find that some pitches work well together, with one setting up the next. This can be considered a proof of concept; the same sequencing can be extended to sequences of three and four pitches for more granularity (at the expense of sample sizes).

To put together these sequences, I first needed pitch-by-pitch data. PitchFX is great for that, if you can get your hands on it. I set up a database of PitchFX records from 2015 to today (and counting!) using the MLBAM Gameday records stored in XML for what seems like perpetuity. From there, finding pitches that end plate attempts is fairly straightforward, as is finding the pitch that preceded it. Because my database only goes back two and a half seasons, data is a little limited, especially for the youngest members of the Orioles rotation.

Consider Dylan Bundy, whose slider is identified as a high-quality pitch by BrooksBaseball.net, noting its high swing-and-miss rate. Fangraphs gives Bundy's slider a much higher value rating than any of his other pitches (his next highest rated pitch per Fangraphs' weighted values is his Changeup). Bundy's slider appears most effective when paired with his four-seam fastball given the triple-slash outcomes of pitch sequences involving a slider (and that end within one pitch of a slider being thrown):
Limited to the 10 most frequent two-pitch combinations to end plate attempts
Bundy has used a four-seam and slider combo nearly 50 times in some order, and the results have been great. Batters are hitting 0.100 off of either pitch when paired with the other. The SL-FF combination produces a slightly higher on-base percentage, which leads me to believe that the fastball is occasionally thrown after the slider goes for a ball and results in some walks.

However, Bundy's slider isn't entirely unhittable. When he goes back-to-back with sliders and the second one ends a plate attempt, they're getting hit well. Batters are slashing roughly .350/.400/.400 when putting a second consecutive slider in play. While this doesn't account for all the times a SL-SL sequence did not result in the plate appearance being terminated, it definitely looks like Bundy should be cautious to throw his slider multiple times in a row - which makes sense, and isn't unique to Dylan Bundy. A batter is expected to get better at identifying a pitch and tracking its movement the more often he sees it.

Kevin Gausman likes to throw heat - and he's good at it. But with sequencing in mind, Gausman might find that the pitch he throws most isn't his best pitch, and that his most frequent pitch sequence isn't doing him any favors. Gausman throws four-seam fastballs over 60% of the time, but they tend to get hit well when batters see them back-to-back. Gausman's split-finger fastball, however, is frequently involved in positive outcomes.

The FS-FS sequence to end a plate attempt has yielded a .180/.260/.270 slash line since the beginning of the 2015 season, and a FF-FS or FS-FF sequence has resulted in a lower batting average and on-base percentage than the more common FF-FF sequence. If Gausman wants to throw hard, and it looks like he does, the pitching staff should encourage him to mix more split-finger fastballs to the mix.

Gausman's full stats show that he throws the FF-FF combination significantly more often than any other pitch sequence, which we should expect given how frequently he throws the four-seam fastball. The positive outcomes on a FS-FS combination come on just 65 instances of that sequence ending a plate attempt, and may be subject to a small sample size. That number of plate attempts is far fewer than the point at which most pitching statistics stabilize. Since the outcomes are so positive, it probably wouldn't hurt for Gausman to test throwing FS-FS sequences more frequently, or just to throw his split-finger fastball instead of his four-seam fastball in some instances.

This type of sequencing can be extended to more pitches to map the value of mixing pitches within a plate attempt, or it can be used to find two-pitch sequences throughout plate attempts that don't result in outcomes. For instance, Gausman's FF-FF sequence might not have the best outcomes when a batter connects with the pitch, but it might coerce more swings and misses or swings outside the zone during an at bat. At the very least, it's helpful for pitchers, catchers, and coaches to understand the value and risk in throwing specific pitches back to back.

1 comment:

Ace said...

Excellent Article! I constantly wonder about pitch sequencing; glad to see it being tracked. A recent example that came to mind was the Sunday game against the Angles. Tied game in the 8th inning, 2 outs, runners on 1st and Second. Here is the Givens pitch selection to Maybin:

- Fastball (strike)
- Fastball (strike
- Slider away (ball)
- Fastball (foul)
- Fastball (base hit, go ahead run scores)

Why on earth would the Orioles not try for the slider away on the 1-2 count? Its a must strikeout situation...throw the pitch most likely to generate a whiff. The batter had clearly timed the fastball by fouling the previous one away.

I think stats like this are crucial going forward. That was a must win game and poor pitch sequencing essentially ended the Orioles season.