/cdn.vox-cdn.com/uploads/chorus_image/image/58573849/843071036.jpg.0.jpg)
You’ve heard a version of this quote hundreds of times: “pitching is about keeping hitters off-balance.” That’s true. And one of the neat things about the era of tools like pitchf/x and Trackman is that now we have data that can help us understand exactly how pitchers do that. A key part of the puzzle is a concept you may or may not have heard of before: pitch tunneling.
The basic concepts are intuitive — you don’t need to be a stats guru to get it. Here’s Greg Maddux’s (brother of new Cardinal pitching coach Mike Maddux) summary of it:
“[My] main goal was to make all of my pitches look like a column of milk coming toward home plate. Every pitch should look as close to every other as possible, all part of that ‘column of milk’. ... Deny the batter as much information—speed or type of last-instant deviation—until it is almost too late.”
Here’s how I conceptualize it, personally: there are three crucial points on the flight path of any pitch. Point A is the pitcher’s release point. Point C is the point where the ball crosses the plate (or enters the batter’s contact zone, which is often a bit out in front of the plate). In between is Point B: the last possible moment where the hitter can decide to stop his swing. If a pitcher can make his pitches hard to distinguish between Point A and Point B, but with significant differences that emerge between Point B and Point C, then hitters will feel unable to trust their own judgment about where the pitch is actually headed. If two pitches do indeed look very much alike up to the batter’s decision point, then they are said to share the same “tunnel” — the “tunnel” is simply the path, from the hitter’s perspective, between Points A and B.
This isn’t a new idea — Joe Schwarz wrote about it from time to time during his time at VEB and continues to at other outlets, if you’re interested in reading more about it. What’s new now is the type of data we have available to examine how well a pitcher uses tunneling.
Previously, the best tool we had was unmodified pitchf/x data. In a groundbreaking article published about a year ago at Baseball Prospectus, Harry Pavlidis, Jonathan Judge, and Jeff Long laid out the concepts and explained their methods for constructing certain stats. In short, they consulted academic researchers and biophysicists to determine where an average hitter’s ultimate swing decision point (what I called “Point B” above) is; reconstructed the flight paths of all tracked pitches with pitchf/x data; mapped release point, decision point, and point of crossing the plate onto those flight paths; and published the results. If a pitcher has a relatively small average gap between release points and a relatively small average gap between decision points, but a relatively larger average gap between where the ball crosses the plates, he’s probably tunneling well.
There was a problem with this, though: what Pavlidis, Judge, and Long were mapping was all viewed from the catcher’s perspective. That’s because, if you’re familiar with pitchf/x data, that’s how it’s presented:
While pitch tracks mapped from the catcher’s perspective would be helpful, what we really care about is what the hitter sees. What the hitter sees — and cannot see — is the reason tunneling works in the first place. As the authors of that BP article describe, by the time the ball is 20-something feet from the plate, it is too late for the hitter’s eyes, brain, and hands to work together to stop a swing. He can’t even do much to change where he’s swinging; by the decision point, the human brain’s ability to track a projectile at 80-100 mph is basically exhausted, and the swing (if there’s a swing) is simply aimed at the spot where the hitter’s brain has already told him the ball is going to end up. Of course, if his brain got it wrong, you get this:
Recognizing as much, last week Long, Pavlidis, and Martin Alonso published an update on Baseball Prospectus (which, to their great credit, must have required an enormous amount of human and computational work). In it, they described how they adapted the previously used data to give a reasonable estimate of how every tracked pitch actually looked to the batter at the plate (and yes, they accounted for batter height and stance; it’s that thorough).
When you look at the results, it’s clear that moving things to the hitter’s perspective matters a lot. Just look at how the exact same sequence of Chris Sale pitches looks to a lefty vs. a righty:
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/10156285/Version_1_768x384.png)
You should really go read the entire thing yourself — this far along in the sabermetrics era, truly breaking new ground is rare and deeply cool — but the upshot is, BP now offers these new stats, both for every tracked pitch pair, and individual pitcher averages:
- Release Distance: how far apart two back-to-back pitches are released out of a pitcher’s hand (i.e. Point A, as I described it above).
- Pre-tunnel Max Distance: the perceived distance, from the batter’s perspective, between back-to-back pitches at the decision point (i.e. Point B).
- Plate Distance: the distance between back-to-back pitches at the plate (i.e. Point C).
- Pre-tunnel Max Time: somewhat convoluted in explanation, but essentially, a lower PreMaxTime means pitches separate later in their flight paths, which means a pitcher is drilling the ball through a narrower tunnel, which is likely good.
- Flight Time Differential: what it sounds like; it describes the change in speed between two back-to-back pitches.
- Plate:PreMax Ratio: maybe the most important one; describes the degree to which difference in position at the plate is greater or less than we would expect from difference in position at the decision point (11.9 was average last year). “Late” break is not a physical possibility in a Newtonian universe, but to an extent “late” break means a pitch that is breaking hard, period, this is where it would show up.
*****
Okay, nearly a thousand words just describing the thing, and I promised this would be Cardinals-centric. So... I’ll try, without exhausting your attention.
It’s important to note that we don’t yet know what a lot of this means, for certain. The data’s only been out in the wild for a week for people to play with; I only read about it over the weekend. It’s most likely that, just as with hitters, there’s more than one way to be a good pitcher, so we shouldn’t read too much (necessarily) into a guy being weak or strong in any of these new metrics. But because these absolutely do tickle the part of my brain that wants to wildly theorize and speculate, here’s how the projected 2018 rotation (sans Miles Mikolas, who has no data and is replaced by Alex Reyes with 2016 data) fared last year, including percentile* ranks (minimum 100 pitch-pairs per side):
*Percentiles reflect judgment calls about what’s “good” — for some numbers I think a smaller number is probably better in a vacuum, for others a larger number. Huge caveats about how preliminary this is, though, and remember that these numbers can and will simply represent differences in approach, not ability, in many cases.
2018 Rotation - Pitch Tunneling Data
Player | RelDist | PreMax | PreMaxTime | PlateDist | FTimeDiff | PlatePreRatio |
---|---|---|---|---|---|---|
Player | RelDist | PreMax | PreMaxTime | PlateDist | FTimeDiff | PlatePreRatio |
C Martinez - vs L | 3.1 (18%) | 1.5 (56%) | 0.167 (68%) | 18.22 (49%) | 0.0232 (38%) | 12.2 (59%) |
C Martinez - vs R | 3.02 (20%) | 1.47 (66%) | 0.169 (50%) | 17.6 (30%) | 0.0251 (49%) | 12 (51%) |
M Wacha - vs L | 2.85 (27%) | 1.64 (23%) | 0.17 (38%) | 18.71 (66%) | 0.0285 (66%) | 11.4 (27%) |
M Wacha - vs R | 3.01 (20%) | 1.65 (21%) | 0.172 (24%) | 17.69 (33%) | 0.0318 (79%) | 10.7 (10%) |
L Weaver - vs L | 1.81 (91%) | 1.48 (61%) | 0.169 (50%) | 17.25 (21%) | 0.0259 (54%) | 11.6 (33%) |
L Weaver - vs R | 1.88 (86%) | 1.31 (96%) | 0.168 (65%) | 15.94 (3%) | 0.0248 (46%) | 12.2 (60%) |
A Wainwright - vs L | 2.2 (62%) | 1.69 (14%) | 0.184 (1%) | 19.28 (81%) | 0.0524 (100%) | 11.4 (28%) |
A Wainwright - vs R | 2.17 (64%) | 1.88 (2%) | 0.18 (1%) | 17.7 (33%) | 0.0471 (99%) | 9.4 (1%) |
A Reyes 2016 - vs L | 1.69 (96%) | 1.42 (80%) | 0.169 (50%) | 16.74 (10%) | 0.026 (54%) | 11.8 (42%) |
A Reyes 2016 - vs R | 1.93 (82%) | 1.7 (13%) | 0.169 (50%) | 19.79 (91%) | 0.0247 (46%) | 11.6 (33%) |
So there’s a big data dump. Are there takeaways? Sure, yes, but I think we should be very careful of overinterpreting this stuff, for now — as I noted above, there is an open question of approach vs. ability in all of these numbers. For example, given that Alex Reyes appears quite good at tunneling to lefties (80th percentile in PreMax), why does he appear bad at doing it to righties (13th percentile in PreMax)? I don’t know for sure, but one ready explanation is that his high PlateDist and low PreMax to righties suggests he’s not trying to tunnel them — instead he’s changing eye level a lot with his power fastball. Which absolutely can work, when you have his fastball.
Still, there are some interesting takeaways. Every time you hear an analyst call Luke Weaver “deceptive,” this helps explain why: he’s got a very consistent release point, and appears very good at tunneling. Why is Adam Wainwright so damn hittable? Because his pitches separate visually at a point when the hitter still has time left to change his mind (1st percentile in PreMaxTime). Is Carlos Martinez an elite pitch tunneler? No, not really — he’s good at it, and perhaps that helps explain why he’s been a good contact manager overall in his career, but he’s not out there imitating Greg Maddux.
There’s a ton of fertile soil left, here. I didn’t even dip into isolating particular pitch-type pairs (FB followed by SL, etc.), and instead cited only the average generic data for each pitcher. It’s possible, even likely, that a pitcher like Martinez could be elite at tunneling particular pitch pairs, while taking an overall approach that minimizes how important tunneling looks for him in general. It’s possible, even likely, that there are things to unearth here that will help explain sudden drops or increases in performance, or suggest ways guys should change. With the ability to understand average leaguewide outcomes on certain pitch pairings — warning: never take analysis seriously if they don’t give you the league average as a point of comparison — from the hitter’s perspective, we can drill down into a lot of cool pitch sequencing stuff we couldn’t before.
There’s a lot left to find out. This is a really neat set of data.