Re-Contexualizing SwStr% for Efficiency
At the beginning of last season, I contextualized the swinging strike rate (SwStr%) (and refreshed those numbers after the season concluded). I had seen other analysts call certain pitches “above-average,” “below-average,” “elite,” etc. using the league-average whiff rate as a baseline. This is neither a criticism nor a judgment, as I absolutely did this before I had my statistically-driven epiphany. But understanding the average four-seamer’s or slider’s or cutter’s whiff rate lends additional context to any assertion one might make about the “elite-ness” of a pitch.
More recently, I wanted to convert discrete outcomes by pitch type into fielding independent pitching (FIP) statistics — namely, FIP and xFIP (expected FIP, which substitutes a pitcher’s rate of home runs per fly ball for the league-average rate). Let me warn you now: the results are very imperfect. It took some brute force on my part to get there, but I got there. I would wager that the the extreme (lowest and highest) values are probably a bit exaggerated. Regardless, it’s an interesting table to ingest:
| Pitch | FIP | xFIP |
|---|---|---|
| Change | 3.98 | 3.99 |
| Curve | 2.11 | 1.93 |
| Cutter | 4.29 | 4.15 |
| Fourseam | 5.19 | 5.15 |
| Sinker | 5.03 | 4.91 |
| Slider | 2.33 | 2.32 |
| Splitter | 2.62 | 2.28 |
| League | 4.15 | 4.15 |
Click headers to sort!
Again, exaggerated values and all that, but you can see just how bad fastballs (four-seamers and sinkers) are relative to breaking pitches (curves, sliders, etc.). What caught me by surprise — maybe not on first glance for you, someone who doesn’t spend way too much time with this data — is that change-ups graded out so poorly. Slightly above-average, sure, but consider: change-ups sported the 3rd-best whiff rate in 2018 (behind sliders and splitters) alongside a coin-flip ground ball rate (GB%). A league-average change-up has 16% SwStr and 50% GB rates, which sounds like an awesome pitch to me. What gives?
Turns out, certain pitch types are more efficacious than others in converting swinging strikes into strikeouts. The league whiff rate (10.7%) is almost exactly half that of the league strikeout rate (K%) of 22.3%, which makes “SwStr% times two” an easy rule of thumb. (I outline that rule of thumb here, in which I compare the efficacy of swinging strikes inside and outside the zone and find that, generally speaking, a pitcher’s strikeout rate should be roughly double his whiff rate.) That rule of thumb, however, does not apply uniformly to each pitch:
| Pitch | K/SwStr* |
|---|---|
| Change | 36% |
| Curve | 58% |
| Cutter | 48% |
| Fourseam | 53% |
| Sinker | 57% |
| Slider | 51% |
| Splitter | 51% |
| Average | 49% |
Click headers to sort!
(Edit, 7:18 pm ET: K/SwStr is calculated using the raw numbers of each — that is, the number of strikeouts divided by the number of swinging strikes. In effect, it’s like saying, “out of 100 swinging strikes incurred on change-ups, 36 resulted in strikeouts on average.” So if a pitcher had a 10% swinging strike rate on his change-up, and he threw 500 change-ups, you could expect, on average, 18 to 19 strikeouts as a result [500 pitches * 10% SwStr * 36% K/SwStr]. It’s a little more abstract, since we’re so used to discussing plate appearance, not number of pitches thrown, as the lowest common denominator. More information, as elucidated by a tangible example of using this calculation, can be seen in the comments.)
Change-ups are actually incredibly inefficient at converting swinging strikes to strikeouts. It must be a function of usage/sequencing, as change-ups lag consistently in this regard. This list…
pitches that ranked in the 90th (*95th) percentile of both SwStr% and GB% in 2018 (min. 200 thrown), with usage in parentheses
anibal sanchez CH (32%!!)
carlos carrasco CH* (16%)
kenta maeda CH* (16%)
noah syndergaard CH* (16%)
stephen strasburg CH (24%)yep — all change-ups
— Alex "Oxlade" Chamberlain (@DolphHauldhagen) January 8, 2019
…feels less impressive knowing those change-ups aren’t converting strikeouts as frequently as I might’ve once expected them to. (Author’s note: the pitch usage numbers up there are a little off, as corrected here. Sanchez is pretty far off. My bad.) It’s not bad at all to have an excellent change-up — a league-average change-up is still significantly better than an elite sinker on the basis of strikeouts — but one should mentally downgrade a pitcher’s seemingly elite change-up, especially when comparing it to breaking balls or off-speed pitches of seemingly similar caliber (i.e., comparable peripherals).
Strikeouts-per-swinging strike is an odd quirk, one I’ll monitor when considering a pitcher’s strikeout rate on a pitch-specific basis. The typical fastball might over-perform the “SwStr% times two” rule of thumb, whereas it’s highly likely the typical change-up significantly underperforms it.
That’s it. No greater takeaway here. Just something to bear in mind while you analyze pitchers ahead of the 2019 season.
RotoGraphs is teaming up with the NFBC to provide unique content throughout the season. To find out about the various NFBC events like the Online Championship go over to the NFBC website to see their various league options.
I’m not surprised by this finding about changeups. The phrase “pitching backwards” almost exclusively refers to throwing changeups in fastball counts. Fastball counts are mostly not two-strike counts. So the sample of changeups thrown is heavily infected by hitters trying to ambush 2-0 fastballs. Hence, lots of whiffs but few Ks.
Glad it conforms to intuition – I’ve reached a point where I assume we’ve optimized beyond traditional sequencing approaches (a good whiff pitch is a good whiff pitch, so why not use it in a whiff situation), but some habits die hard, I guess.
Honestly, this is a little hard to follow (for me at least).
When you discuss “K/SwStr”, are you talking about a pitch-specific K%/pitch-specific SwStr%? If so, then I’d expect a ratio more around 200% (K=SwStr x 2).
Or are you talking about absolute #s? Total Ks occurring for on pitch/Total swinging strikes for that pitch
A swinging strikeout is ultimately related to how often a batter whiffs with 2 strikes. If the batter has 2 strikes, then a swinging-strike fast ball is no more efficient that a swinging-strike change-up. Both are 100% effective. You are no doubt looking at all strike counts when calculating K/SwStr of 36% to 58%, which might suggest that pitch utilization/selection is the reason that change-ups appear less efficient in generating Ks.
There are at least 2 things (that I can think of) which might make a 2 strike change-up less efficient on a per pitch basis:
1. Perhaps the SwStr% of a change-up goes down when their are 2 strikes (due to a batter being more defensive and better at fighting off change-ups). For example, the SwStr% goes from 18% with 0-1 strikes to 14% with 2 strikes. In this case, a change-up could be less efficient in generating strikeouts.
2. Perhaps the change-up is equally effective in capturing swinging 3rd strikes, but less effective on a per pitch basis of capturing looking 3rd strikes. In the case of Anibel Sanchez, his change-up has the lowest zone percentage of any of his pitches (32% last year), possibly making it his least efficient pitch for generating strikeout looking.
I am sorry for the long post. If I am the only one who is struggling with this, please down-vote! 🙂
No worries! Part of me had this feeling I didn’t explain part of it well enough, but couldn’t pinpoint exactly what. This was it.
The rule of thumb “SwStr% times two” and K/SwStr are using completely different sets of denominators. The rule of thumb relates to SwStr% to K%, the former of which is calculated as a percentage of all pitches and the latter of which is calculated as a percentage of all plate appearances. That they are about double (or half) of the other is just a pleasant coincidence. K/SwStr, on the other hand, uses the raw number of swinging strikes as the denominator rather than SwStr%. So, for K/SwStr for change-ups, roughly every third swinging strike ends in a strikeout. The calculations might look as follows:
Rule of thumb: 10% SwStr * 2 = 20% K
K/SwStr: 10% SwStr on 1,000 pitches thrown = 100 swinging strikes. A 50% K/SwStr means 50 K’s from those 100 whiffs. Say each PA lasts 4 pitches, so 1,000 pitches / 4 per PA = 25 batters faced. That’s 50 K’s in 250 PA – aka, 50/250 = 20% K. A little more complex, but gets us to basically the exact same spot. It’ll just be different for each pitch type (and for each pitcher depending on his own per-PA or per-game efficiency).
I imagine the discrepancy specifically for change-ups relates to usage. I don’t think effectiveness changes by count; just that change-ups are simply thown less often in advantageous counts. (Brad Johnson alluded to this in his comment.)
Anyway, hope this clears things up a bit.
Thanks. On the surface, the strikeout efficiency chart seems a little counter intuitive. But your explanation and reasoning make a lot of sense. Great article.
I’m glad you asked because it definitely needs the clarification. I’ll make a note in the post!
Hi Alex
Interesting post.
I got confused by the Pitch type FIPs chart, where the mean for each is 4.15, but the values for XFIp are lower for all but Change-ups. Since this is statistically impossible, unless 80% of all pitches were changeups, what gives here?
Are the means not actually the same?
FIPs and xFIPs are higher for four-seamers and sinkers (you’ll see if you look again). That probably resolves your confusion; the other component is not all pitches are thrown with the same frequency, so a simple raw average of the FIP/xFIP values won’t produce the league-average value. But fastballs make up a disproportionate chunk of all pitches thrown, so they skew the league-average toward them despite pretty much every other pitch being better than league-average.
Hey Alex… I like the work, thank you! I’m curious if you could take this a step further and normalize the data based on count? Fastballs are thrown more when behind in the count—when hitters have the greatest success so I’m thinking that fastballs might be showing a higher FIP than they deserve considering the situation. Could we see all pitches within the same count or at least pick a neutral count and only examine the performance of the pitches in that count? I’m also curious about 2 strike counts since O-swing% goes up. Thanks!