Pitching mechanics, the uncertainty of data, and fear

In Paul Nyman’s first article for The Hardball Times, he wrote:

Barry Zito’s lost fastball became my bridge too far.

That’s how I felt in attempting to write this, my first article for The Hardball Times. Every time I tried to attack it, I was repulsed by credibility issues. I could not, in a single article, deploy the necessary reserves of information and knowledge that I believed necessary to successfully achieve my objective.

My 15 years of experience dealing with how the body throws the baseball kept telling me I was reaching beyond what is accepted as “good pitching mechanics knowledge.” Every attempt to achieve my objective was blocked by entrenched pitching mechanics ignorance and culture. The amount of reader preparation necessary to establish overwhelming credibility would push the supply lines of throwing mechanics information far beyond the breaking point and doom the article to failure.

Paul wrote this nearly two years ago, about the time I was starting my baseball training business——Driveline Baseball —after writing for some time on pitching mechanics at my now-defunct SB Nation blog, Driveline Mechanics. I had a lot of perceived disagreements with Paul at the time, much of it dealing with injury risks to pitchers and how he seemingly ignored these problems. Over time, I learned to appreciate his contribution to the field of throwing mechanics and training baseball pitchers, and it costs me nothing to say that he was right about a great many things and that I was wrong about a lot of those issues as well.

Paul defined the “bridge too far” as trying to attack the concepts of pitching/throwing mechanics in a single article, and until I sat down to write this introductory article for The Hardball Times, I couldn’t fully appreciate it. I’ve been sitting on this article for months now—just ask Studes!—and I’m happy to finally put some words down to hopefully outline some of the problems we face when we talk about analyzing pitching mechanics.

Defining the Problems

Before we can attack the issues with developing pitchers, mitigating injury risk, and increasing release velocities of prospective pitchers, we have to figure out where the problems exist. In my mind, there are two major problems with developing pitchers at all levels of baseball (from Little League up to MLB):

1) Inability to cheaply and accurately collect the appropriate data regarding “pitching mechanics”
2) Fear due to misunderstanding of “newer” training concepts and methodologies

I am very happy to say that “we” (the baseball public community) are making gigantic strides with regard to the first concept. The second one… not so much. But let’s see what we’re doing so far!

Data Collection: What Cannot be Measured Cannot be Improved Upon

Strangely enough, the concept of accurately measuring something and incrementing it over time is something that people lose sight of when it comes to training baseball pitchers.

We live in a society where we have limitless access to information from reputable sources—Academic Earth —offers full courses from prestigious universities like MIT, Yale, and Harvard; these courses are accessible to anyone in the world with an Internet connection! MLB Advanced Media publishes their PITCHf/x data freely, and writers and analysts all over the Internet have taken to deconstructing the data in the pursuit of knowledge. My company—Driveline Baseball —has made huge strides in developing a low-cost biomechanical analysis laboratory (under $1,000 when comparable setups are well over $15,000), and we plan on publishing that data for open analysis.

And yet… when I bring up the concepts of measuring release velocities, kinematics, kinetics, and training loads over time for a baseball pitcher (particularly youth athletes), I am often greeted with fear of the unknown. Skepticism is a very good thing—you should be very wary of anyone trying to sell you on a “new” concept— but being closed-minded is not. Fortunately for me, I’m stubborn, so it doesn’t bother me all that much!

There are two major pitching-related data collection efforts going on today in their relative infancies—PITCHf/x and Field Biomechanics. I am heavily involved in the latter and slightly involved in the former, and I believe both are very important to the future of pitching development. Let’s take a look at where we are in both fields.

Data Collection: PITCHf/x

Most everyone knows of the major PITCHf/x databases out there – TexasLeaguers, BrooksBaseball, and the father of them all, Josh Kalk’s old tool. There are proprietary systems out there, like Mike Fast’s database that he uses for analysis at Baseball Prospectus. I have put together a PITCHf/x database myself and combined it with historical injury information, available at the Advanced Baseball Injury Database page. These are all great resources and definitely help to shine a light on the dark confines of pitching mechanics. But for all of the glory, there are some serious issues with them all.

Uncertainty of Data

It seems to me that Colin Wyers (Director of Statistical Operations at Baseball Prospectus) is leading the latest charge of skepticism on this front, and I could not be more thrilled with this. For all of the successes of sabermetrics over the past decade (led mostly by the publication of Moneyball), we have taken many steps backwards with the popularization of metrics that may not be very accurate. We champion defensive statistics like UZR that bases its algorithm on data collected from stringers employed by Baseball Information Solutions, but we do not question these subjective measurements of data! I could write thousands of words on this topic, but instead I’ll just point you to an excellent thread on TangoTiger’s Book Blog to read—the indeterminate quality of data. Most everything in that thread revolves around BIS pitched ball data, but it can also absolutely be applied to PITCHf/x data.

A Hardball Times Update
Goodbye for now.

Analysts like Mike Fast and Josh Kalk have written correction algorithms to help fix the calibration errors between parks, but this does not solve the very big problem in how this data is displayed: Metrics like fastball velocity, release point (x/z planes), and spin deflection are issued verbatim. This is a huge problem. As physicist Walter Lewin (MIT) said:

Any measurement that you make without knowledge of its uncertainty is completely meaningless.

Or, if you prefer, Richard Feynman:

I know how hard it is to know something.

I’m not going to pontificate too much more on this, as my point should be clear: PITCHf/x data (to date), while very interesting, is riddled with errors that we probably cannot permanently fix (we do not know the uncertainty). No open source correction methods exist at the time of the writing of this article, and I’m putting a call out for people to help write this algorithm. If you’re interested, please get in touch with me – contact details are at the end of this article.

Data Collection: Field Biomechanics

About the concept of keeping video archives of their pitchers, Paul Nyman once said this:

Recently I was asked by an MLB team’s baseball operations person to look at one of its pitchers, a player who last year was consistently 92-94 mph and who this year is throwing in the 86-88 mph range. My first question: Do you have good video of this player? The answer was no; they had had what commercial television footage was available. I then lectured this person on the necessity to create and maintain a player video library where camera angles are carefully chosen and the videos maintained to be used in situations like this.

The clips they sent me and those I was able to find online showed almost no difference in how the player threw the baseball, which is more typical than not for high-level performers such as major league pitchers.

For years, Paul stressed the need for video from these angles but also tempered this with the fact that 30 FPS video could only tell you so much about a pitcher’s delivery over time.

Let me say this: There are “reputable” people out there—affiliated with big names—who offer “biomechanical analysis” packages where you film yourself pitching from various angles. You then submit this film to the company and they perform their “biomechanical analysis” on the data. This is all done with 30 FPS-quality video.

In my opinion, such two-dimensional analysis is limited at best. The calculations are useful to a degree, but it is hardly advanced analysis. That said, I am fully aware of how you can do these types of analysis—by filming the pitcher from the open face side (third base for a righty, first base for a lefty), you can (sort of) pick out the point of shoulder maximum external rotation (MER) and measure the angle. If you include a control object (like a yard stick or baseball bat of known size), you can also figure out stride length, stride deflection, shoulder abduction and other two-dimensional values as well by filming from behind the plate. Some companies charge upwards of $300 for the privilege of having a protractor applied to a monitor and recommendations made based on a database of “elite” major league pitchers’ measurements.

If you think this type of analysis is of value, I have good news for you: In the coming six weeks, I will show you how you can perform two-dimensional planar biomechanical analysis by yourself using free and open source tools. This is the first phase of the Open Biomechanics Project, and we will talk more about this as my series of articles at THT goes on.


But if two-dimensional biomechanical analysis isn’t the answer, then what is?

Three-Dimensional Biomechanics: Direct Linear Transformation

Direct Linear Transformation (DLT) is the process by which multiple two-dimensional images are combined to create a three-dimensional representation of the model. The area that the two-dimensional images are capturing must be defined by a control object – an object of known size and proportions. For example, here is my control object that I built at the Driveline Biomechanics Research laboratory:


DLT relies on this control object to know where things are in space and time, then uses it as a reference to deconstruct and recombine the synchronized two-dimensional images into a three-dimensional model. Suddenly, from 2 or more cameras (we use 3), you have this:


(graph is (Shoulder ER-90deg) vs. ball velocity (m/s))

If you’re a quick thinker, then you’ve already beaten me to this thought: We have PITCHf/x cameras installed in every stadium, capturing the flight of every ball thrown by the pitcher. Why not install high-speed cameras in every stadium and calibrate them using a control object? We could have digitized biomechanical information not on every pitcher – but on every pitch thrown by every pitcher in any stadium!

We could use this data to analyze a pitcher’s biomechanics as the season wears on, as they tire throughout a game, and to compare populations against one another.

Technology was simply not there for Paul’s dream to come true—but we are getting closer every day. I have worked tirelessly over the past four years to bring a low-cost system like this to fruition, and with the help of many, many people*, we’re starting to make some serious headway. My research assistant, Matthew Wagshol, and I are getting very close to producing reliable kinematic reports of pitchers similar in quality to labs that cost upwards of $15,000. These labs require pitchers to come into their facility, put on a bunch of biomechanical markers, and throw off a fake mound in sneakers into a net. What we’d capture would be the real deal—game situation biomechanics!

In many respects, this is what Will Carroll has been pining after for years. Today, we have the low-cost technology to make it happen for teams of all sizes – not just MLB teams, but minor league affiliates, colleges, and even high schools.

Moving Forward: Addressing the Fear

The next article will outline “the fear” that people have when it comes to training athletes, and how it is both detrimental to performance of major league players and how it substantially increases their injury risk.

We need to embrace technology—skeptically, of course!—to move forward. We must understand that what someone did to get to the major leagues is often what will keep them there – we should not fear things like Trevor Bauer’s unorthodox mechanics and training protocol. We should be amazed by them, and we should investigate them.

I’m not picking up where Paul left off—I would never claim to be the heir to his pile of research and experimentation that he’s done. I’m simply adding to the pile in hopes that we are permitted to even ask the necessary questions to improve pitching performance and hopefully mitigate injury risks that plagues our sport we love so much.

References & Resources
* Special thanks to Richard Betzel, who helped me for years with the concept of applying DLT to baseball pitching

Kyle owns Driveline Baseball and Driveline Biomechanics Research, and has authored The Dynamic Pitcher, a comprehensive book and video set dedicated to developing elite youth baseball pitchers. He is also a consultant for an MLB team and a major Division-I college program. Follow him on Twitter @drivelinebases or email him here.
Newest Most Voted
Inline Feedbacks
View all comments
Sir Larry
13 years ago

This looks appears to be an exciting path you are heading down, Kyle.  Nice work, thus far.

As both an avid fan of the statistical and mechanical side of the game, I am very intrigued at the potential performance/injury assessments you are building towards with advanced uses of Pitch F/X data.  As an amateur level coach of 16-21 year old players, I would definitely have interest in the Open Biomechanics Project for its ability to reduce injuries and improve better long-term habits.

Mike Fast
13 years ago

Nice article, Kyle!  I would love to have data on pitchers’ mechanics from high-speed video that could be related to PITCHf/x data.

Speaking of which, are you familiar with the Scientific Baseball guys?  Doc Schoenhals and Fred Vint were in San Francisco at the last PITCHf/x summit.

Kyle Boddy
13 years ago


Thank you. I’m in the middle of typing it all up, but there will be a website talking about the Open Biomechanics Project. We need to make this type of basic technology and analysis available to coaches and parents everywhere – and not charge them $300-400 for the privilege.


While I don’t (yet!) have a contract with a professional team, I plan on taking a lot of high-speed video this summer of PacNW pitchers to make available on my site. With any luck, we’ll be able to set up a control object on the mound, film it, and do some trial field work outside of our lab. If I get this stuff, I’ll be sure to make it available to those interested.

Not familiar with the Scientific Baseball guys. Their website isn’t very exhaustive.

Kyle Boddy
13 years ago

Dr. Nathan:

Thank you for the kind words.

1) ASMI’s system is different than what used to be the industry standard – manual digitization of subjects from multiple cameras using DLT. Many research papers outline this methodology. Werner et al.‘s paper “Relationship between throwing mechanics and elbow valgus in professional baseball pitchers” outlines it perfectly.

By using some similar methodologies, you could theoretically use neural net processing to give you a basic “map” of the pitcher and tweak it manually by hiring data entry people. Or you could leverage a system like Amazon’s Mechanical Turk to “automate” it. There are a lot of options – tracking every single pitch may not be something that can be feasibly done in the near future, but certainly capturing a series of pitches for each pitcher per game could be done.

2) Interesting. I commented further on Tango’s blog post about this article; I think using multiple points of z=0 and not a symmetrical model lends itself to serious calibration problems.

I love the physicists’ phrase “It is precise, but not accurate.”

Alan Nathan
13 years ago

Kyle, several comments on your very nice article:

1.  As you may know, the American Sports Medicine Institute (http://www.asmi.com), led by Dr. Glenn Fleisig, has been using high-speed motion analysis to study pitchers for years.  The major focus is injury prevention (Dr. James Andrews, the famous surgeon, is a major player in ASMI).  They have a multi-camera system (I think at least 5 but I don’t recall offhand).  The system uses infrared cameras and requires a reflecting material that is attached to whatever it is you are trying to track (e.g., various places on the pitcher).  Typically little styrofoam reflecting balls are velcro’d to the pitcher.  Lighting is always an issue with these types of systems, making it difficult to employ it in a game situation. 

2.  A variation of DLT is exactly what PITCHf/x uses to get 3D coordinates from 2D images from two cameras.  There are two separate issues:  precision and accuracy.  The precision of the measurements has been pretty well established.  It is more difficult to characterize the accuracy.