Is The Fastball Less Valuable Than We Thought?
We’re coming off a season in which the American League Cy Young winner had a meh fastball. It’s true. Corey Kluber‘s sinker didn’t meet the above-average benchmark for swinging strikes (5.6%, his was 5.3%) or grounders (50.5%, his was 48.6%). Look at National League breakout Tanner Roark’s two-seamer, and you’ll see the same thing. Is it possible that we’ve overrated the importance of the fastball with respect to overall outcomes?
Fastball velocity is usually the first thing on the scouting report, and with good reason. We know that every mile per hour is worth a fraction of a run allowed per nine innings. So it seems important.
I asked Steven Staudemeyer to run a correlation between the swinging strike rates on the four major pitch types and a pitcher’s overall swinging strike rate. Here is that table (all p values are less than .0001).
| Pitch | Correlation (r ) | r^2 | Prevalence |
|---|---|---|---|
| FF | 0.678 | 0.460 | 57.7% |
| SL | 0.495 | 0.245 | 13.7% |
| CH | 0.484 | 0.234 | 10.4% |
| CU | 0.301 | 0.091 | 9.9% |
Obviously the fastball is important. It’s thrown nearly three-fifths of the time! The fastball’s swinging strike rate explains almost half the variance in the pitcher’s overall swinging strike rate.
But look at the slider and the change. They each explain nearly double the variance in swinging strike rate than you’d expect if you just went by volume alone. It makes intuitive sense — you use the fastball to get ahead, and you use the other pitches to finish them off — but it has all sorts of practical applications.
Take, for example, a pitcher like Dan Straily. His fastball was down below 90 mph before he was traded to the Cubs, so he had below-average velocity for a right-hander. He still managed above-average whiffs (7%), but few would say that his best asset is his four-seamer. But if you look at his change (18% whiffs) and slider (21.7% whiffs), you see that he has two pitches that break in different directions and get good results. You might take a chance on him if you thought that he could use the fastball mostly to get ahead, and if you thought the fastball was a little bit less important than it seemed by volume.
There are other names on this list, too. Ivan Nova has a nice curve (13%) and slider (23%) and an iffy four-seamer. He could switch to the sinker and have more success, even if it’s mediocre. Teammate Wade Miley doesn’t get 50% grounders with his sinker or 6% whiffs with his four-seam, but he does have a nice change (14.4%), slider (18.4%), and curve (10.5%) combo. Jerome Williams keeps getting shots despite his fastball because his change (16.5%), slider (16.7%), and cutter (10.8%) all rate as above-average by whiffs.
On the other hand, this is clearly just the first step. Each pitch is interrelated, and so there are issues with comparing these correlations. And there are probably diminishing returns on throwing off-speed pitches more, so volume is still really important. But there has to be a way to study this further. Ideas are welcome.
With a phone full of pictures of pitchers' fingers, strange beers, and his two toddler sons, Eno Sarris can be found at the ballpark or a brewery most days. Read him here, writing about the A's or Giants at The Athletic, or about beer at October. Follow him on Twitter @enosarris if you can handle the sandwiches and inanity.
I wonder how this changes when you control for count. Hitters are more likely to whiff on two-strike pitches than 0-0 pitches (source), so what if you did the same sort of analysis for each of the 12 ball-strike states?
There’s only 11 right?
0-0 (1)
0-1(2)
1-0 (3)
1-1 (4)
1-2 (5)
2-0 (6)
2-1 (7)
2-2 (8)
3-0 (9)
3-1 (10)
3-2 (11)
But that’s beyond the point. Yes it would be interesting to have whiff rates for each count.
Never mind, I missed 0-2
I think the way to study this further is with interaction terms. You could do a hypothesis generating multiple linear regression with all the pitches plus their interactions. You could get really crazy and do 3-way interactions too, but that might get pretty complicated to interpret. Not to mention somehow working pitch sequencing into that, which is kind of the elephant in the room of pitching analysis that no one has really done much with yet.
Also, are swinging strikes really the thing you want to estimate? Sliders are generally pitches designed to get a swing and miss, while CB and CH more mess with a hitter’s timing and are probably taken for strikes more often. Swinging strikes can be a good proxy for stuff, but I’m not sure it’s telling the whole story, and there might be a more direct measure than that.
I think pitch sequencing would fit into a regression pretty nicely if the response variable was the binomial whiff outcome of the pitch. The variables on each pitch would include categorical variables for last pitch, and second-to-last pitch, or something like that.
Controlling for count could be done using a simple (strikes – balls) with a categorical variable for two-strike counts to simplify the 12 counts a little more.
In fact, to your point about outcome, you could make a continuous outcome for the pitch result’s value. Each value could be the pitch result’s linear weight or whatever.
I don’t really want to use ball in play data, other than maybe GBs. Focusing on whiffs allows us to use clean, non-BABIP numbers.
I like the start, Eno, and I always enjoy your work.
That said, this is going to be a very challenging problem, and I’m not sure this or any variation of this, or any similar methodology will give us the answers we want.
For one thing, four seam fastballs are used to do more than just generate whiffs and grounders. They can also induce pop-ups and low fly ball authority if used well up in the zone. The fastball is also generally the easiest pitch to throw for strikes/throw where you want it for a lot of (most, probably) guys. Strikes (taken, swinging, foul balls) and location are important independent of the pitch’s type and character.
But the benefit of this kind of study, I think, would be in how to identify possible diamonds in the rough by picking out guys with poor fastballs but decent off-speed and breaking pitches who could overachieve just by throwing fewer of their worst pitch. To use it as a blanket assessment, though, I think would be problematic and not especially insightful.
It’s just so hard to handle an issue as complicated as pitching and generate a simple yet comprehensive model, I guess.
I think called strikes, swinging strikes, GBs and IFFBs might be the ‘cleanest’ data to use.
Wade Miley isn’t teammates with Nova. Vidal Nuno now is. Brad screwed this up in his McCarthy article today.
Is there some kind of collective Fangraphs blackout regarding the return on the McCarthy trade? An error in the Matrix.
I think the big issue is that fastballs are used less in high leverage situations, like with 2 strikes, especially by pitchers like Kluber.
I haven’t seen much work done which works on the leverage of different counts. I think this is a direction which either needs to be re-emphasized or further researched.
I think there’s often a faulty assumption built into models which assumes that a 0-0 count is just as important as a 3-2 pitch. Obviously, it isn’t.
The days before the spin rate analysis…