September 21, 2019

Statistically Speaking: To Pull Or Not To Pull Your Starter

Among the many on-the-job lessons Matt Williams is learning in his first season as Washington Nationals’ manager has been the fine art of knowing when to pull a starting pitcher, due to ineffectiveness or fatigue. It’s an elusive skill and between it and bullpen management, can make or break a team’s season and a manager’s career. Unfortunately for many, it is a skill that is more art than science, with a lot of trial and error involved in the process.

For Williams, there has been lots of error in his trials thus far. But recently it appears a corner has been possibly turned, with his handling of Doug Fister’s most recent outing. Fister, only a handful of starts into his 2014 season after suffering a right lat strain, had been a mixed bag results-wise in his first three starts, and after 83 pitches in the sixth inning May 25th against the Pittsburgh Pirates, Williams had seen enough, despite going well over 100 pitches in his previous two starts and pitching well enough against the Pirates to get the win. However, vigilance surrounding the now-healthy lat ruled the day, and the righthander was relieved by Craig Stammen, an ever-lowering and inconsistent release point—possibly a result of fatigue—the culprit behind the early curtain for Fister.

Here’s a look at Fister’s release points for the 83 pitches, broken down by pitch type and inning, with the last pitch that sealed his fate circled in red. As noted in a previous article, x0 is the horizontal component of the release point (also labeled HRel) and z0 is the vertical component (VRel):

Fister RelPt

It was a savvy move by a green manager, who up until this point, had all too often been the victim of his moves backfiring; it appeared that Williams’ handling of his pitching staff was taking a turn for the better. However, looking at the release point graph, it appears that the incriminating pitch wasn’t necessarily that bad and there were plenty of other pitches that reflected a dropping release point that is often a red flag for fatigue. Yet, it does provide an interesting jumping off point for what the threshold could be for pulling a starter due to fatigue, with the caveat that game situations can dictate whether a pitcher gets pulled immediately upon exhibiting fatigue or if they stay in to finish an at bat.

Using this 83rd pitch as the threshold pitch, let’s calculate some z-scores in an effort to standardize the release point data—both HRel and VRel individually—so that we can compare all pitches, in light of the fact that each pitcher will have a slightly different average release point for each of their pitch types. The closer a z-score is to zero, the closer a pitch is to his average release point, with large negative values consistent with a drop in arm angle and release point. Calculating the respective z-scores across pitch type for Fister and then plotting these values across pitch count and inning gives us the following results: first, HRel data:

Fister_HRel

I have done the liberty of drawing the line across the data (arrow pointing at pitch 83) as a reference to this fatigue threshold; pitches below the line can be considered worse pitches from a release point perspective.

Here’s the same graph for z-scores of the vertical component of Fister’s release point:

Fister_VRel

It appears the vertical aspect has more ‘danger zone’ pitches, with a large number of four-seam fastballs (FF) seen, reflecting some inconsistencies with getting on top of the pitch, especially later in the game. VRel data also has larger negative values associated with it compared to HRel, which could possibly lend it being more affected by fatigue than HRel.

Overall with Fister, we do see a very cautious approach taken by Williams—Fister spent most of the game with z-scores at or above average, with only a handful of pitches going past the threshold of pitch 83.

Let’s go back to the previous day’s game, with Stephen Strasburg on the hill. Unfortunately for Strasburg, he ran into a little more trouble over the course of the last inning he pitched and there are many who felt that Williams kept him in too long. With a pitch count of 91 after completing six innings—but laboring through a 22-pitch sixth inning in the process—and surrendering a solitary run up to this point, Strasburg sputtered in the seventh, tallying another 17 pitches and giving up two runs on the way to a loss at the hands of the Pirates.

Like we did for Fister, let’s look at his raw release point data across pitch type and inning, with his last, ‘red zone’ pitch circled:

Strasburg RelPt

While the situation is a little different here compared to Fister, in that Strasburg was allowed to finish the inning and did not get the hook as abruptly as Fister, we do see some similarities between the pitches. With Strasburg, we again see the last pitch not being egregiously erratic with respect to the release point, but we do see some steady drop in arm slot in previous innings along with some inconsistent arm slots in the final frame, a possible clue that Strasburg was fatigued before the seventh inning.

Let’s move on to z-score data, again starting with HRel values:

Strasburg_HRel

…and the VRel values:

Strasburg_VRel

The data are slightly more damning compared to Fister’s; there are a large number of sub-threshold values in both horizontal and vertical components of release point; also Strasburg essentially spent the last two innings in this release point ‘danger zone’, potentially indicative of the 22- and 17-pitch innings he finished his outing with were not only stressful, but inconsistent due to arm fatigue.

The number of pitches under the threshold, across pitch type and release point component, are provided below in order to compare the duo:

Pitch Type HRel VRel
Fister CH 1 3
CU 2 3
FC 1 3
FF 7 9
FT 2 3
Strasburg CH 11 17
CU 6 12
FF 18 28
FT 1 3

CH: changeup, CU: curveball, FC: cut fastball, FF: four-seam fastball, FT: two-seam fastball

It is again clear that Strasburg’s outing, one where he battled hard and still put up a respectable pitching line, was less than ideal, compared to Fister’s. This is also an interesting contrast in how Williams handled his staff in the span of a day—one pitcher, despite some apparent clues that he was not long for his outing, is left to pitch through some bumps in the road, while another, who despite recently coming off of injury, still appeared to have a little more left in the tank, is quickly taken out of the game.

While not all variables have been taken into consideration—injury status, game situation, bullpen status, among many others—this exercise does potentially provide a very rough method of not only monitoring fatigue (albeit retrospectively), but also gauging a manager’s tendencies with how he handles his pitching staff as a whole and as individuals when they begin to labor in their outings.

***

Data courtesy of Brooks Baseball.
____________________

Stuart Wallace is a Contributor to District Sports Page. A neuroscientist by day, the Nevada native also moonlights as an Associate Managing Editor for Beyond the Box Score, stats intern at Baseball Prospectus, and a contributor at Camden Depot. A former pitcher, his brief career is sadly highlighted by giving up a lot of home runs to former National Johnny Estrada. You can follow him on Twitter @TClippardsSpecs.

About Stuart Wallace

Stuart Wallace is a Contributor to District Sports Page. A neuroscientist by day, the Nevada native also moonlights as an Associate Managing Editor for Beyond the Box Score, stats intern at Baseball Prospectus, and a contributor at Camden Depot. A former pitcher, his brief career is sadly highlighted by giving up a lot of home runs to former National Johnny Estrada. You can follow Stu on Twitter @TClippardsSpecs.

%d bloggers like this: