THE BOOK--Playing The Percentages In Baseball

Looking at his player card:

http://baseball.bornbybits.com/2008/Barry_Zito.html

So far this season Zito’s FB has been at 84.78 MPH, so close to the Fangraphs numbers. Could just be a calibration error between parks, though.

MGL,

I’ll try to answer some of the questions posed.  The radar guns at stadiums, for TV telecasts, at least when my father was the Phillies producer, were manned by reliable production assistants.  Not to say these people committed numerous errors but there are many variables that must be exactly right for a radar gunner to get the absolute correct measurement.  This measurement from the guy in the stands is then transmitted to the stadium scoreboard or wherever it is… a place like ESPN, from what I’ve heard, will train a camera on that scoreboard so when we see the score strip on TV display velocity, we’re seeing a transmitted result from a reliable crew member in the stand manning a radar gun.

BIS uses the scorestrip on TV telecasts, so when you see Zito throw an 85 mph fastball appear next to the 3-2 score, or whatever it is, that is what BIS records.  From speaking to several BIS representatives, they understand it isn’t the end-all method for measuring velocity but their clients are MUCH more interested in classification and location.

Fangraphs uses BIS data, which is why it goes back as far as it does.  Now, because of this there may be some discrepancies between what gets classified as what, as BIS and Pitch F/X use different methods (human vs. robot).  So, for something like what Fangraphs reports for Zito as compared to Josh’s cards, I would guess some combo of park calibrations and classification discrepancies… Zito throws slow to begin with so it wouldn’t be out of line to suggest some of us are classifying changeups as heaters or vice versa.

The Pitch F/X data (which yeah is the same thing as Gameday, from MLB Advanced Media) records both Start Speed and End Speed, and it’s been speculated before that the End Speed will hold up more from park to park as compared to the Start Speed. 

I’ll check Zito shortly unless someone beats me to it.

According to what I have, Zito’s last three starts:

86.24
85.60
85.71

So if Josh’s cards have his overall averages at 84.78, then it would seem he is throwing faster recently.

Eric covered a lot of the questions well already, but I’ll fill in a little from what I know from Sportvision and the PITCHf/x side.

The pitch speed given by PITCHf/x (since about mid-2007) is measured 50 feet from the point of home plate.  They chose this distance because it best correlated with radar gun readings on average.  They originally had reported at 55 feet in order to get the top speed close to the release point but were reporting speeds above that on the typical radar gun, so they switched to 50 feet to match the radar guns.

As Eric said, the radar gun reading is very dependent on the operator.  The radar gun gets radar returns from everything that is moving--the pitcher’s arm, the pitcher’s hand, the baseball--and those speeds change over time.  So getting a consistent speed with a radar gun is a difficult enterprise.  I’m sure some operators are better at it than others.  You can’t just look for the peak speed because that is the pitcher’s hand after releasing the ball.  (The preceding paragraph is my understanding based on conversation with Ken Milnes at Sportvision, who is their radar expert.)

A well-calibrated PITCHf/x system should be spot-on with the speed, much more accurate and reliable than a radar gun, within about 0.5 mph of random measurement error on a given pitch.  I can go into the details of how the system was calibrated to check the pitch speeds if anyone cares.  The details are in Marv White’s presentation from the PITCHf/x summit.

However, there are sometimes issues with camera calibrations--spherical distortion coefficients, position of the camera, etc.--and these can introduce some errors ballpark-to-ballpark (or occasionally in-season in the same park) that make the initial speed measurements (at 50 feet from the plate) be off by a couple mph.  Those are things that you can figure out if you look at the data closely, but you have to be aware of what to look for. 

The plate speed measurements from PITCHf/x are pretty reliable even in the face of most of the camera calibration issues because there are a lot of calibration points right around home plate.  Those of you familiar with photogrammetry will understand how that helps improve the accuracy.  Basically, interpolated points have less error than extrapolated points.

My rule of thumb is that pitches lose about 10% of their speed over 50 feet.

As far as Zito’s fastball speed specifically, I’d have to look at his data more closely.

Zito’s fastball speed according PITCHf/x has been mostly in the 81-85 mph range this year.  In his last three starts he’s been throwing his fastball in the 84-87 mph range.

Chan Ho Park’s fastball speed has been roughly in the 90-95 mph range most of the year, and he was about 89-94 mph in his last start.

I haven’t audited any of the above data to check for problems with camera calibrations, but I think the conclusion that Zito is throwing a little faster in his last three starts is probably sound.

Mike - Although the Sportvision camera’s track each pitch from near the pitching mound circle all the way to the batter, the 9 parameters that determine the equations for the trajectory of the flight of the pitch are determined from the ball positions that are in defined “regions of interest”.  These regions of interest are defined differently for each field, but usually start about 35 to 40 feet from home plate and stop 8 to 12 feet from home.  This is done to exclude areas that would have movement from the pitcher, batter, bat, coaches, runners or fielders that would confuse the ball recognition portion of the Pitch f/x program. 

So, although your above explanation above is correct that there are more points of calibration closer to home plate and that makes the ball locations closer to home slightly more accurate and less sensitive to camera abberations, in fact, the locations and velocities at both the pitcher end of the pitch and the home plate end of the pitch are extrapolated data because all the curve fitting is done to points between about 10 and 40 feet.

Your post could be read that there is a possibility of a 1 to 2 MPH bias of initial pitch speed between some home fields.  I was under the impression that this year any error of that magnitude was transitory and quickly fixed by the Pitch f/x operator and would be unlikely to last an entire game.  Have you found evidence for persistent differences between parks this year?

Peter, thanks for your clarification.

I know MLBAM would like the impression to be that any errors are transitory and quickly fixed; that is not the reality.  I’m sure Sportvision would readily admit that.

Cincinnati has had significant errors in the x-acceleration all season that continue to this day (last I checked was July 2, I think).  Detroit had errors earlier in the season that lasted a couple weeks or so.  There are other things going on at Philly, San Diego, etc., that I have not characterized well yet, but are definitely persistent over a period of games and perhaps the whole season.  Sportvision is well aware of this and working on determining the exact problems with the camera calibrations.  I’m probably not at liberty to share any more than that at this time.

However, it’s also worth pointing out again, as Marv White said at the summit, improving on the current level of data quality is a bit like asking the singing cat if he can also play the piano.  What we have from PITCHf/x, both in terms of data quality and the sheer amount of information available, is much better than the quality from any other data source, be it Questec, BIS, or someone else.