Pitch Type Performance: 2018 Summary
Shortly after the onset of last season, I dug into pitch-level statistics to see how much swinging strike rate (SwStr%), ground ball rate (GB%), and isolated power (ISO) varied by pitch type. I felt inspired after analyzing Madison Bumgarner before the 2018 season and noticed his fastball, once elite, was utterly broken after his dirt bike accident. (See his 2018 player caption and this July post in which I followed up MadBum’s lack of progress.) I felt encouraged by the praise the post received from readers and fellow analysts alike for the clarity it provided. I’d like to think it helped move the needle, even if only slightly, in terms of how we evaluate pitchers.
I wanted to refresh the guts of that post for the 2018 season with additional metrics. There’s not much else to discuss; this’ll be short and sweet. (I’ll toss in some gratuitous high-level analysis following these tables.)
Notes:
- All data is courtesy of PITCHf/x via Baseball Prospectus
- All tables present average rates for starting pitchers only
- Due to pitch tracking/stringing not being perfectly precise, the numbers below are highly accurate but not completely so and may not align exactly with FanGraphs’ batted ball data (for example, Baseball Info Solution strings far fewer line drives than does PITCHf/x)
- Click headers to sort!
Batted ball outcomes by pitch:
| Pitch | GB% | LD% | OFFB% | PU% | HR/OFFB | BABIP |
|---|---|---|---|---|---|---|
| Change | 50% | 23% | 21% | 6% | 19% | 0.272 |
| Curve | 49% | 24% | 22% | 5% | 20% | 0.300 |
| Cutter | 43% | 27% | 22% | 8% | 20% | 0.294 |
| Fourseam | 34% | 27% | 29% | 10% | 19% | 0.295 |
| Sinker | 52% | 24% | 19% | 5% | 20% | 0.310 |
| Slider | 44% | 24% | 23% | 8% | 19% | 0.280 |
| Splitter | 53% | 23% | 19% | 5% | 22% | 0.275 |
Hitter production allowed by pitch:
| Pitch | SwStr% | AVG | SLG | ISO | TAv | wOBA |
|---|---|---|---|---|---|---|
| Change | 16% | 0.237 | 0.387 | 0.150 | 0.241 | 0.292 |
| Curve | 13% | 0.221 | 0.366 | 0.145 | 0.223 | 0.267 |
| Cutter | 12% | 0.255 | 0.417 | 0.162 | 0.258 | 0.315 |
| Fourseam | 9% | 0.263 | 0.460 | 0.197 | 0.282 | 0.349 |
| Sinker | 6% | 0.292 | 0.455 | 0.163 | 0.287 | 0.355 |
| Slider | 17% | 0.208 | 0.348 | 0.140 | 0.217 | 0.260 |
| Splitter | 18% | 0.209 | 0.342 | 0.133 | 0.219 | 0.262 |
Due to limitations with the data, wOBA is approximated (but still highly accurate)
Quick recap:
- Sliders, change-ups, and curves are good. Four-seamers and sinkers are bad. Splitters are an enigma!
- Seriously, don’t throw a sinker.
- No, I’m kidding a little — throw a sinker, but only if it’s actually good. Power sinkers preferred. (Just discussed this regarding Chris Archer recently.)
- Most pitchers could improve by optimizing their repertoires to include more breaking stuff in place of fastballs, or to more frequently lean on, say, a two-seamer in lieu of a four-seamer.
- That said, the way a pitcher’s offerings interact is important.
- Lastly: I’ve seen a lot of analysis using pitch values recently. Pitch values are the ERA of pitch-level analysis, as it’s entirely outcome-driven. Does ERA tell the full story? Never. It can be truthful, yes. However, almost every baseball fan has already learned to approach ERA with a healthy amount of skepticism. Do the same with pitch values. Use the information above to further refine your understanding of pitch values as a description of a pitch’s effectiveness.
Good stuff Alex.
What do you mean by “power” sinker? Is there a certain minimum velo threshold?
Glad you asked! This is something I’m working on — a “what makes a great [pitch type]” style of series — but it’s a fledgling endeavor so I don’t have a solid answer. I will say, anecdotally, among pitchers from 2014-18 with 500+ sinkers in a season, ~60% were “above-average” with velo 94+ compared to ~40% with velo 92-93.
All of the results of pitches, for individual pitchers and collectively, appear to be available, except for foul balls and balls. I would think that this could be a crucial omission in the data. Except when there are two strikes, foul balls are just as good to a pitcher as a swinging strike, perhaps a tiny bit better because you cannot allow a stolen base on them. For example, if sinkers or four-seam fastballs generate higher foul ball rates and lower ball rates than other pitches, it could make sense to use them early in counts—they are generally the easiest pitches to throw for strikes (which I am sure is why pitchers continue to use them disproportionately to their outcomes) I would think, and getting ahead in the count is better than getting behind.
One pitcher comes to mind here—Folty. It has been mentioned numerous times at FG that his K% was not sustainable because his swinging strike percentage was too low. But watching him, his four-seamer seems to generate an inordinate amount of foul balls (and he continues to throw it with two strikes, which tends to jack up his pitch count). If that is true, perhaps that is how he (and possibly others) can end up with K%’s higher than their swinging strike percentage would indicate.
I imagine sequencing is also at play…when pitchers are behind in the count, fastballs of all types are pretty likely but also more likely to get hit.
I would agree that sequencing is the largest determinant of variance in pitch value/performance, both in-season and relative to other pitches of that classification. One complicating factor is pitchers can’t control when hitters swing at certain pitches (although they can do their best to influence the hitter’s decision). So sequencing can only account for so much, in addition to so many other luck-based factors that seem to plague pitchers. But, in short: agreed.
1) [edit] wOBA doesn’t account for balls and foul balls directly, as it is a product of PA outcomes. But balls and foul balls contribute to PA outcomes. What good do foul balls do if the pitch gets torched on contact? [/edit]
2a) It was me who torched Folty in a recent xK% post.
2b) In 2018, Folty’s four-seamer was only 20th percentile among all four-seamers in terms of foul ball percentage.
Fair enough on 2b, I was just thinking anecdotally. As for 1, I don’t get your point. The good foul balls do if the pitch gets torched on fair contact can be explained by the following example (I know the numbers are not realistic, they just demonstrate the point:
A pitcher has a four-seamer that has a 80 percent foul rate, 15 percent ball rate, and 5 percent fair-contact rate, with the results being abysmal. Throwing it early in the count leads to two-strike counts very often, when the pitcher uses breaking stuff to strike batters out. Batters are less likely to chase breaking stuff out of the zone when 2-0 in the count than 0-2, so the four-seamer has helped the K% without actually doing any of the K’s itself.
Or more succinctly: strikes 1 and 2 are meaningful, too.
Ah, if you’re concerned primarily with the value added per pitch: pitch values might be what you’re looking for. It assigns a value for every pitch in every plate appearance. Accordingly it operates more closely to a pitch “ERA” whereas everything I’ve highlighted (except SwStr%) focuses exclusively on the final pitch (outcome) of the PA.
The pitcher you described would be a good pitcher, I think, despite his bad fastball because of his usage/sequencing patterns. However I think there’s too strong an assumption here that foul balls indicate effectiveness. There are really good fastballs with low levels of foul balls and high levels of foul balls. There’s a spectrum of relationships, and I wouldn’t characterize their effects as linear.
Would it be reasonable to say that foul ball rate allowed with 0 or 1 strikes = good, and foul ball rate allowed with 2 strikes = bad?
Out of curiosity, why didn’t you use xwOBA of any other xStats?
Because, ideally, xwOBA will converge on wOBA for the league over a full season. Take sinkers, for example, which have a 0.344 wOBA… and an exact-same 0.344 xwOBA. Slider is off by nine points(Also, note that the wOBA value differs from my post because I used PITCHf/x in lieu of Statcast, which has a million pitch type classifications. A 0.344 wOBA or a 0.355 wOBA are both really, really bad.)
it would be interesting to compare this to how every pitch type fairs in a 1-1 count.
League wOBA (xwOBA) on all counts: 0.315 (0.311)
League wOBA (xwOBA) on 1-1 count: 0.370 (0.369)
That’s a good spoiler for how much worse every pitch is on 1-1. It’s ugly.
Great article. Have you done any pitch performance analysis separating performance to same handed/opposite handed batters? Would be interesting to know how the average slider thrown by a LHP performs against lefty vs righty hitters.
No, unfortunately the PITCHf/x data I use isn’t that granular, but you (or I or anyone) could use Statcast data to check this using its database search function. The results will likely be different since pitch types, balls in play, and pretty much everything are coded/stringed differently, but different results are better than no results. Here’s an example for sliders in 2018:
RHP vs LHH: .281 wOBA, .268 xwOBA
RHP vs RHH: .258 wOBA, .252 xwOBA
Interesting! Not as big of a split as I would have expected.