Designated Hitter January 19, 2006
A Quantitative Approach to Studying Release Point Consistency

We know an awful lot about pitchers. We know how hard they throw, how many batters they strike out, what kinds of pitches they have, and whether their deliveries are fluid and easy or violent and rough. This is all objective and indisputable information that has a lot of value when it comes to projecting a pitcher's future health and success.

One thing we don't know much about, though, is the consistency of a pitcher's release point. The fact that we don't have a good way of measuring what's arguably the most important part of being a good pitcher is one of the more ironic twists of modern analysis. Sure, you can look at a bad curveball and say "he let go too early" or "he held on too long," but that's just one of a few thousand pitches that the guy's going to throw all year, so it doesn't tell you very much. What we need is a way to quantify the extent to which release points varies over a larger period of time for different pitchers. And, thanks to MLB.tv and MS Paint, we're getting there.

Take this low-quality screengrab of Felix Hernandez, immediately prior to release:

The ball is a little fuzzy, but it's definitely there about to leave his hand. Just moments from now, it's going to fly past some unfortunate hitter's bat at 95 miles per hour, since all Felix ever does is dominate. But anyway, here's the really cool part: copy and paste that picture into MS Paint. Now move your cursor to the center of the ball. In the lower right corner of the window, there should be a set of coordinates - for me, it reads 124,37. Think of this like a set of coordinates on any generic x,y plot. The center of the ball is 124 units (pixels) from the left of the window, and 37 from the top.

I didn't know quite what to make of this the first time I noticed it, but after a little brainstorming, I realized that this could be an effective way to quantify both release point location and, with a large enough sample, consistency (it's the second one that I actually care about). So I devised a plan: collect a group of images of a pitcher much like the one of Felix above, enter the x,y location of the ball into a spreadsheet, and calculate 95% confidence intervals at the end to get an idea of his release point consistency.

For the purposes of this article, I decided to compare Mark Prior to Kerry Wood, since one is considered to have picture-perfect mechanics while the other...not so much. As far as further methodology is concerned, note that:

1) For each pitcher, I looked at 40 pitches - 20 from the windup, and 20 from the stretch. These are kept separate, in case either pitcher happens to change his delivery with runners on.

2) I only used images from one game so that I didn't have to account for differences in center field camera angle. Incidentally, both Wood and Prior's games took place on the same day - April 13th, a doubleheader vs. San Diego.

3) All pitches were chosen randomly.

4) To account for any differences in scale between images (since the camera has a tendency to zoom in and out), I chose reference points at opposite corners of the batter's box and adjusted accordingly.

5) Once I had 95% confidence intervals in pixels, I converted to inches by using the fact that a baseball is about three inches in diameter, and showed up as eight pixels wide on screen.

So, onto the results of this study:

Kerry Wood, Windup

The little red box represents the 95% confidence interval - based on the collected data, Wood would be expected to release 95% of his pitches from the windup within a box measuring 1.6 by 3.1 inches. The area of this box is 4.84 square inches.

Kerry Wood, Stretch

Same deal - based on the data, Wood should throw 95% of his pitches from the stretch within a box whose dimensions are 2.5 by 2.8. The area of this box is 6.95 square inches.

Although we obviously don't have a baseline for how much variation you'll see in a standard pitcher's release point, since this is a fledgling analysis, Wood's results compare favorably to Felix's, at least as far as pitching from the windup is concerned.

Mark Prior, Windup

Dimensions of the box: 3.4 by 5.0. Area: 17.0 square inches. That's more than three times the variability that we saw in Kerry Wood's release point, which I wasn't expecting. That's a huge difference.

Mark Prior, Stretch

That's more like it. Dimensions of the box: 1.6 by 2.3. Area: 3.52 square inches. Better than Wood, and better than Felix.

Based on the results of this study, you'd be tempted to say that, as long as there isn't anyone on base, Kerry Wood's release point is significantly more consistent than Mark Prior's. I don't think anyone expected that to be the case ahead of time, but that's what the numbers bear out. To get a better indication of whether or not this is really true, though, you'd have to compare them over more than one game. I would've loved to do just that, if only the whole process weren't so time-consuming - it took me six hours to look at Joel Pineiro over three starts by himself, so even accounting for the fact that it gets quicker as I become more familiar with the procedure, comparing two pitchers is at least an entire day's work. So, consider this as more of an introduction to the methodology rather than a comprehensive investigation.

As far as Prior is concerned, something that may have influenced his final numbers is the fact that the game I looked at was his season debut, six days after throwing 87 pitches in a minor league rehab start. Looking at the spreadsheet, Prior's release point was getting lower as the game wore on, the gradual dropping of his arm being a possible sign of fatigue with his arm strength not yet at 100%. It is worth noting, though, that his considerable improvement with men on base in the April 13th game wasn't a fluke - opponents put up a .766 OPS against him with the bases empty last season, striking out in 23.1% of their plate appearances, but with men on their OPS dropped to .578 while their strikeout rate jumped to 32.8%. In 2005, Mark Prior was a much better pitcher with men on base, and based on the results of this preliminary study, it may have been because he was way more consistent with his release.

A study like this is going to have both its strengths and limitations. On the downside, it's a very tedious process, as you're going to lose the better part of an entire day if you're looking to gather any sort of meaningful results. It's also strictly two-dimensional, as it doesn't give a real good idea of how far forward the ball is being released. Short-arming the ball a foot in front of your throwing shoulder is going to make it do one thing, while a full-extension release will make it do quite another. The system could be improved by having a side-view camera providing z-axis location information, but until then we'll have to maximize the resources at hand. Additionally, it should be noted that this kind of investigation would be difficult to perform on someone who deliberately mixes up his arm angles, a la Jamie Moyer.

All that said, the biggest issue is how to properly interpret the data once you have it. Release point consistency is nice and all, but a consistently good release point and a consistently bad release point are two very different things that cause very different things to happen. Looking at the numbers, we can see that Kerry Wood's release point was more consistent with nobody on than Prior's, but Wood got slapped around while Prior was terrific. Why? Was Wood releasing the ball in a bad place all game long? I'm guessing that better pitchers will generally have more consistent release points, and that these kinds of apparent exceptions are considerably outnumbered by their opposites, but I can't say that with any degree of certainty.

Accepting that any new kind of study is going to have its kinks to smooth out, I think the potential upside here makes it worth pursuing further. Although it's possible to perform manually, one could conceivably automate the whole process, turning what used to be three hours of work into three minutes of watching a machine do everything for you. With the data that would provide, you could look at anything - release point consistency against right- and left-handed batters, correlations with things like walks and strikeouts...pretty much everything you can do with the stats we already have. Comparing a guy's consistency when throwing different pitches (fastballs, curveballs, changeups, etc.) could prove to be a pretty telling indicator of what he needs to work on in the bullpen. And those are just a few examples. Like with any metric, you could use this one in any number of ways.

Still, the most important thing here is: it's something. It's putting a number to what used to be educated guesswork. And as far as I'm concerned, doing that is always worth the effort.

Jeff Sullivan is the creator and primary author of the Lookout Landing Seattle Mariners blog. He's also a student at Trinity College, although nobody's sure where he finds the time. He can be reached by email at jeff@lookoutlanding.com.

Bravo! I've been reading Jeff's work over at Lookout Landing and Leone For Third for a long time. He's a great writer and I'm glad to see his work getting some love around the blogosphere. This stuff is fascinating, Jeff. Keep up the good work!

Wow, this is great stuff. Keep it up.

perhaps prior's greater variability in release point is a *good* thing, compared to wood's. The reason being is that you have to (at least, i'm pretty sure you have to) release different pitches at different points. Perhaps Wood releases his curve and his fastball in the same spot, diminishing the effectiveness of one of the pitches. Prior has different release points for each type of pitch, which keeps the batter off balance and allows each type of pitch to be thrown reliably for a strike. I think the release point for each type of pitch needs to be analyzed. And perhaps a consistent release point isn't all that great, even for the same pitch. It keeps batters off balance. Batters don't watch the pitcher, they watch a virtual square where they know the ball is coming from. If a pitcher suddenly releases from somewhere else, it can throw everything off.

not to say this isnt an interesting idea though. its a good start. didn't mean to be all negative. 8)

Kevin, I think that's only true if the player is doing it purposely. Part of the reason that a pitcher like Orlando Hernandez is so successful is that throughout a game he uses about a dozen different arm angles. But, there aren't a ton of pitchers that can get away with doing this.

As a Cubs fan, I don't think Prior is a pitcher that intentionally changes his arm angles. If a player doesn't do that, the next most important thing is repeating your delivery. Jeff's study indicates that for at least one start, Mike Marshall may be right that Prior's delivery is not so perfect.

And thanks are in order to Jeff for such a fun study, and a great start.

I agree that this is fascinating work, and work that needs to be applauded and done.

MLB has installed cameras all over the field, to capture fielder positioning and movement. All that we need is for the brains behind the FoxPuck to put the microchip inside a baseball, so you will know the position of the ball at all times. And take the geniuses of the video games that hook up all the players, and do that during the real games too. (Can video game companies have more cash handy than MLB? It sure seems so.)

I think Jeff did some nice work here, but it's decidedly *not* a good start; it's a restart. This type of work has been done, documented, and published by the American Sports Medicine Institute and others using much better video and technology. At least ten major league teams use a technology called Dartfish that allows them to use any source of tape to look at mechanics, calculate kinetic forces, and document release point changes from pitch to pitch and pitch type to pitch type.

Jeff's attempts at "open sourcing" this info are nice, but ignore the fact that others have done this better elsewhere. He also ignores the research that shows that release point itself is all but irrelevant, rendering any of this useless.

It's good work and interesting, something I don't want to belittle. It's just not useful.

My comment comes off harsher than intended. It's the fact that the work is closed, not published or documented well, and hasn't been addressed by the sabermetric community that led Jeff to this. It *is* good work by Jeff and I think once he gets up to speed, we'll all see more good work.

I'm glad Will retracted the harshness. "It *is* good work by Jeff and I think once he gets up to speed, we'll all see more good work.".

It's not a restart, if it's not widely available. If the ASMI did publish something, can you give us a reference?

Jeff,

I've read you're other stuff regarding release point consistency and I also think it's great. I was a pitcher who struggled greatly with this in college. If you have a good arm, it's probably the difference between a very good pitcher and an erratic one.

Off the top of my head, I can think of 3 ways to make money doing this. The first part is the same:

-Write software that will work with digital video of taken from 2 angles (so you get a 3-D release 'box')of a pitchers deliver that will quantify the results. The software would be so tricky, because how do you tell the computer when the ball is being released? Perhaps when the entire sphere becomes visable? (So many pixels of white in a certain area, I don't know...)

Then...

1. Use at games or practices to write a published study about the importance/effects of release point variability. Probably the worst idea of the 3, but it could lead to the following 2.
2. Bring 'the system' to a team and let them hire you to do the analysis for their own purposes.
3. Sell 'the system' as an out of the box application (DV camera's sold seperately). Probably could include some drills too.

Great work and good luck! I hope it bears fruit for you in some fashion.

Tom - the studies are pretty varied. Anything using the Ariel or Eagle High Speed motion analysis system is fair game here and likely is at least tangentially related. I don't have access to the full studies, but AJSM (American Journal of Sports Medicine) and the Journal of Biomechanics are two main places that Fleisig et al have published.

Andy - all that's been done. (2) is a complete non-starter due to the culture of baseball. Anyone who read 'Moneyball' knows what happens to a good new system. I've been talking to a guy that has proof of how you can beat hitters and what pitches they can't hit, but can barely get teams to listen. He's caught between giving away too much and giving enough to make them listen. It's much the same for this.

One thing I will add is that release point is *completely* irrelevant to results. Tom House has had this in his books for the better part of a decade.

Here's a couple good starter points from ASMI:

http://www.asmi.org/asmiweb/research/usedarticles/biomechanicalfactors.htm

http://www.asmi.org/asmiweb/research/usedarticles/highlowpitches.htm

I can attest to #2 as well. Working for a single-team is not a "money-maker". You have to sell limited services to alot of teams. Teams may generate 130 million\$ each, but they act like they generate 3 million\$. They will scrimp and save on everything, and go overboard on Cristian Guzman.

***

"release point is *completely* irrelevant to results" sounds like a b.s. statement to me. Perhaps to Roger, Pedro, RJ, and Prior it doesn't matter. But for the guys on the bubble, or in the minors?

The one basic truth, to anything, is that EVERYTHING is relevant. The question is: "To what extent is x relevant to y". (Other than benign things like the relevance of how I feel impacting Jessica Alba. As much as I want that to be relevant.)

From Will's link: "Therefore, a pitcher wishing to improve pitch velocity should focus on optimizing the position of the trunk. "

***

There is a distinction between a "release point" based on the whole body being off, or having a release point that you purposely change, while still keeping the rest of your body in synch. It's not necessarily that having an inconsistent release point is bad, but that you *may* infer that the inconsistent release point is due to your body being all over the place.

So, rather than concentrating on the release point, you can also try to concentrate on the pitcher's hips, legs, and arm angle.

Consistency in arm movement and release point is exactly what makes a pitcher with a good fastball and changeup so devastating--think Santana. Coming out of the hand, they look identical to the hitter--except a difference of 20mph.

Thanks for the comments Will and Tango, I've read a lot of your guys' work. As with most things in the market (for anything), if there's a need - you can bet someone is trying to fill it. Especially with the dollar values involved in pro baseball.

A couple of thoughts jump to my mind - using a camera mounted behind the delivery hand I find it difficult to imagine obtaining a precise determination of the true release point. Also as noted in one of the comments (I think someone was gettin at this), I would expect some variation in the release point in an effort to gain some minor deception or change in velocity (i.e., holding th ball slightly longer to slow it down just a tad). Just a thought.

Sorry guys, you're all wrong. We learned this past season that a pitcher's effectiveness is based on arm angles.

you don't hold the ball LONGER to slow it down...you hold it tighter and deeper within the hand.(ie the change-up is held against the palm, where as a fastabll is held away from the palm and with less pressure on the fingers.)

So I went through three year stats on ESPN (2003-2005 totals), and compared OBP for Prior and Woods when no one is on base, and when there's a man on first but no one else on base. Ostensibly, I've isolated windup versus stretch, am I right? Interestingly, Prior has a .048 higher OBP from the windup. But .012 fewer walks/HB per PA. Kerry has a .018 higher OBP from the windup vs. stretch, with .003 more walks/HB per PA.

In both cases, the walks per PA difference seems pretty marginal. I guess you could propose that they're going to pitch more down the middle of the plate with a man on first than with no one on, so you would assume that mechanics aside, there would be significant fewer walks per PA in those situations... Thus, the fact that you see almost no difference could mean that, in fact, both are less accurate from the stretch, countering that intent. But that's guessing that the absence of a difference means there's two countervailing effects, rather than simply assuming there's no difference in accuracy between the stretch and the windup.

But looking at the OBP it's interesting. You would expect to see a lower OBP from the windup, not a higher one, right? I mean, the theory says you use the windup when you can to make it harder for the hitter, but switch to the stretch to prevent base stealing. Maybe it's a situational effect -- they're challenging the batters more with no one on, putting up more hittable pitches?

In any case, I THINK (maybe I should say "guess"?) this supports Jeff's analysis. Namely, Prior does worse from the windup (.302 OBP) than he does from the stretch (.254 w/ man on first only), whereas Woods is pretty much closer to the same performance either way (.321 vs. .303).

It also suggests that maybe Prior should simply pitch from the stretch all the time... But that's probably the tail wagging the dog, I may have the causality all wrong, and there are other factors at play...