The Biggest Pitcher K% Outliers of 2018

Mike Foltynewicz, a first-ballot Hall of Namer, immediately strikes me as someone who outperformed his strikeout rate (K%) in 2018. I don’t have to look far for confirmation: his 27.2% strikeout rate outstripped his 10.3% swinging strike rate (SwStr%) by a mile. Because whiff rate correlates so strongly with strikeout rate, it serves as a useful proxy for what one could expect of a pitcher’s strikeout ability.

I generally follow this rule of thumb when I’m reluctant to get too into the weeds when assessing peripherals: SwStr% * 2 = K%. It’s imperfect but useful in a pinch. Folty violates this rule of thumb pretty dramatically. Of 13 qualified pitchers who struck out at least 27% of hitters, his 10.3% swinging strike rate falls well short of the shortlist’s 2nd-lowest mark (Charlie Morton, 11.9%). Foltynewicz’s 2018 performance has already wilted under what amounts to very little duress.

Still, I wanted to allow Foltynewicz the opportunity to redeem himself. Whiff rate does not a pitcher make; there are other components to plate discipline allowed such as chase rate (O-Swing%) and zone rate (Zone%), among others, that describe each pitcher in much finer detail. I broke down a pitcher’s plate discipline allowed into its component pitch outcomes:

• Swing and miss in the zone: Zone% * Z-Swing% * (1 — Z-Contact%)
• Swing plus contact in the zone: Zone% * Z-Swing% * Z-Contact%
• Swing and miss outside the zone: (1 — Zone%) * O-Swing% * (1 — O-Contact%)
• Swing plus contact outside the zone: (1 — Zone%) * O-Swing% * O-Contact%
• No swing, in the zone: Zone% * (1 — Z-Swing%)
• No swing, outside the zone: (1 — Zone%) * (1 — O-Swing%)

These six outcomes, when expressed as percentages, sum to 100%, comprising all pitch outcomes. I set all six as the independent variables in a regression equation that specifies strikeout rate as the dependent variable. (Technically one of the independent variables drops out of the model due to multicollinearity, which is something I need to state to satiate only the most statistically nerdy readers; everyone else can move onward unfazed.) Using data that spans 355 pitchers across the 2014-18 seasons, the model produces a 0.78 adjusted r². For reference, swinging strike rate alone, when regressed against strikeout rate, produces a 0.72 adjusted r². In other words, employing the full swath of plate discipline components only marginally improves the robustness of the model’s fit. Still, I feel more comfortable trying to describe strikeout rate using all the information about a pitcher’s zone presence rather than one-tenth of it. I would guess that the largest exogenous (external) impacts would relate to sequencing, at this point. (Note: there already exists a pitcher expected strikeout rate (xK%) as formulated by colleague Mike Podhorzer. See my footnote after the table at the end.)

I calculated the difference between the expected and actual strikeout rates for every qualified pitcher from the 2018 season. Here’s every pitcher who overperformed by at least 3 percentage points:

1. J.A. Happ (+6.14%)
2. Gerrit Cole (+4.58%)
3. Rick Porcello (+4.33%)
4. David Price (+4.22%)
5. Justin Verlander (+4.06%)
6. Foltynewicz (+3.68%)
7. Jake Odorizzi (+3.10%)

Ah, there’s our boy, Folty. As expected (or, at least, as I expected), his name floated to the top. His 23.5% expected strikeout rate pushes his 3.77 xFIP (and identical SIERA) north of 4.00. His 2.85 ERA relative to his ERA estimators already signaled good luck; that he shows up as a top strikeout rate overperformer suggests there is even more luck unaccounted for. There are solid skills here, but he’s likely primed for a pretty big letdown relative to his National Fantasy Baseball Championship (NFBC) average draft position (ADP) among the top-25 starting pitchers and top-80 overall. Like, he’s being drafted as a true-talent mid-3.00 ERA pitcher, not a true-talent high-3.00 ERA pitcher and definitely not a potentially true-talent low-4.00 ERA pitcher. He falls into the trap of mid-round starting pitchers that I routinely avoid, so he’ll likely top my shortlist of players to fade.

Happ had, for all intents and purposes, a career year by measure of his peripherals. I guess it shouldn’t come as a surprise that he tops a list during his age-35 “breakout” season. I’m adamant about evaluating pitchers on the basis of their individual pitches (so catch-all posts like these have become increasingly unusual for me). Happ has some decent pitches — his four-seamer and sinker are above-average for fastballs but middling in the grand scheme of things — but the overall picture painted by a career-high 10.4% swinging strike rate suggests he overachieved. I have a cheap share in an ottoneu league on which I might try to sell high this offseason. (There’s precedent for his overperformance — his +2.64% mark in 2015 ranked ninth among 78 qualified starters — which might relate to the relative effectiveness of his fastballs. Still, his +6.14% mark last year was an untenable 2nd-highest overall the last five years.)

The Cole-Verlander and Porcello-Price tandems raise an interesting multi-pronged question: Elite catcher framing? (Max Stassi and Sandy Leon ranked among the top-7 catchers in framing last year, per Baseball Prospectus.) Or high-quality fastballs? Or, maybe a question Trevor Bauer might ask of the Astros: doctoring the ball? (You didn’t hear it from me.) Or is it just noise?

Here’s every pitcher who underperformed by at least 2 percentage points:

1. Luis Castillo (-4.82%)
2. Kevin Gausman (-3.76%)
3. James Shields (-3.42%)
4. Kyle Hendricks (-2.74%)
5. Lucas Giolito (-2.65%)
6. Tyler Anderson (-2.31%)
7. Dylan Bundy (-2.24%)
8. Miles Mikolas (-2.21%)

Before I discuss Castillo, it’s worth revisiting the Cole-Verlander paradox, which I think is less a paradox and more an example of how good and bad pitchers distinguish themselves. The list has a couple of distinct underperformers, but generally speaking, Shields, Giolito, and Anderson are not good pitchers. Some might argue Gausman is, but I wouldn’t. (I wrote about Gausman after the 2017 season, and he deserved it, but it’s before I had attained a finer understanding of the nuances of pitch-level performance.) Same with Bundy, who has one of the game’s best sliders but also one of the game’s worst fastballs. It’s intriguing that Bundy might be a 27% strikeout guy, but he’s a fly ball machine in a bad home park for it. I should note there’s still room here to blame the catchers; Caleb Joseph, Matt Wieters, and Omar Narvaez were decidedly below-average (for the latter, league-worst) framers, per BP.

Castillo is hardly a buy-low — through the early returns of NFBC drafts, he’s the 32nd starting pitcher off the board — but I’ll bet anyone he out-earns Foltynewicz, who’s being drafted almost 40 picks sooner, in 2019. Castillo has his problems — namely, calling Great American Ballpark home — but the peripherals are generally superb for a second-tier arm.

Hendricks trails Castillo by 10 drafts slots overall, so, again, not really a buy-low here. (Y’all are a sharp lot.) Among pitchers who have thrown at least 500 innings since the start of 2014, only Lance Lynn has outperformed his SIERA by a wider margin than Hendricks. At a certain point, you have to chalk it up to skills; he’s a command artist who induces a ton of weak contact. The 2018 season might’ve been his “worst” by peripherals, but he should, and probably will, bounce back. It seems like I’m preaching to the choir, though, given his normal draft price.

I told you to draft Miles Mikolas instead of Luke Weaver last year, so know that I was in on Mikolas from the start. I was less enthused about his strikeout rate, though, which I thought might exceed 20%. (Then again, I didn’t expect a 3.6% walk rate, which more than made up for it.) It’s good to know there’s a little bit of strikeout rate upside to provide cushion for when he inevitably regresses in 2019. He’s not a 2.83 ERA true-talent pitcher, but he’s so strikingly similar to Hendricks, it’s not out of the question to expect Mikolas will simply outperform time and again. At 38 picks prior to Hendricks, though, I will almost certainly opt for the safer clone with the longer track record.

Maybe some of you are wondering (or maybe only I wondered): how well do a pitcher’s plate discipline metrics stick from year to year? Like, the exact breakdown of the six component parts of his plate discipline allowed — those can’t stay exactly the same, can they? Of course not. But a pitcher’s plate discipline allowed in year 1 produces an adjusted r² of 0.42 when regressed against his year 2 strikeout rate. That’s pretty strong, and while it’s a far cry from how well it describes actual (year 1) results, it’s still strong enough, in my opinion, to make this kind of analysis worthwhile.

What I’m sure most of you are actually wondering, though: where’s the full list? Right here, fellas. Note this is qualified pitchers only.

Footnote:

In 2017, Podhorzer updated his expected strikeout rate metric for pitchers that, by measure of adjusted r², is far superior to this. The primary difference is it relies on Baseball Reference data. I used FanGraphs data for the following reasons:

• It’s most familiar to me, and all my other research uses it.
• Podhorzer’s equation, while good, double-counts strikes (as strike percentage, or “Str%”, and again as its component parts of L/Str, S/Str, and F/Str). Doing so might artificially inflate the fit of the model. This is only minimally problematic — it doesn’t make the metric bad, in and of itself — but the high adjusted r² may be misleading. I would recommend calculating, for example, looking strikes (L/Str) as a percentage of all pitches (L/Str * Str%) rather than of all strikes, and then delete Str% from the equation entirely.
• Lastly, the equation doesn’t really draw any influence from the zone itself, and it doesn’t indicate which looking/swinging/foul strikes occurred in or out of the zone. It’s more dependent on outcomes than FanGraphs’ plate discipline data. A swinging strike is a swinging strike, but it gives no indication of whether that pitch should’ve been a ball or strike. A ball is a ball, but it gives no indication of if a pitcher got squeezed. I guess what I’m trying to say, much more succinctly, is that FanGraphs’ plate discipline data gets as close to input-based measurements as possible, compared to Baseball Reference’s outcome-based measurements.