My Two Cents on PITCHf/x

PITCHf/x is a popular graphical analysis tool that offers great accessibility to public domain and a complementary advanced scouting system with user friendly platform which allows Sabermetric enthusiasts to manage and repurpose content and extract valuable insight to give a glimpse into the art of pitching.

Accuracy of PITCHf/x Classification 
With a sophisticated system that is constantly evolving, it is hard to deny PITCHf/x can be a pretty precise and accurate “estimation” of pitch type recognition. Kershaw, Strasburg, Harvey and many other pitchers with distinct repertoire are extremely easy to categorize; but in reality, computer algorithm designed by MLB Advanced Media often has difficulty with pitchers that have a large variety of pitches in the arsenal. 

According to MLB Advanced Media’s classification, Chien Ming Wang didn’t throw any change-up on 2013/6/16, Wang mainly employed split-finger fastball and it accounted for 14.1% of his pitch selection. On the other hand, Pitch Info identified thirteen pitches as change-up and no splitter was recorded. So what did Wang throw?

 

 

The fact is both pitches were thrown on 2013/6/16. Is it objective and conservative to conclude that PITCHf/x classification has plenty of room for improvement?

Trustworthiness of PITCHf/x Movement Analysis 
PICHf/x claims that it has the capacity to measure and quantify pitch movement. As a result, it gives the so called sabermetric community an enlightened gain of ability and perspective to evaluate horizontal and vertical pitch breaks. So do these nice looking charts and sabermetric-like stats really provide a meaningful intuition in the context of pitch movement for any particular pitch or pitcher?

Case-case comparisons of horizontal movement with Darvish’s slider:

Method one
Investigating three sliders with similar pfx_x value on 2013/4/2: A, B and C; with 13.233, 14.045 and 13.204 pfx_x respectively.

Slider A

Slider B

Slider C

Looking at the GIFs, both A and B are hanging sliders, contradictive to what PITCHf/x has suggested. In term of vertical movement, A has –3.502 pfz_z, B has –0.34 pfz_z, and C has a conflicting +0.208 Pfz_z.

Method two
Examining PITCHf/x possible systematic error of measurement. A non random selection approach is employed. The goal is to explore the magnitude of data issues and calibration errors on derived quantities, such as pitch movement.

4.6 inches lateral movement

 7.9 inches lateral movement

The above GIFs suggest that pitch movement by itself doesn’t really paint a picture. The process of finding what attributes go into making a filthy slider from other derivatives  calculated by PITCHf/x requires further interpretation of horizontal and vertical displacement. So it makes sense to briefly look at the relation of pfx_x or pfz_z with other parameters. The finding is quite intuitive. In term of slider, there is a negative correlation between pfx_x and Vx_0 in Darvish’s first ten starts this year. The GIF comparisons also have emphasized the great risk for falsified raw database that can lead to a barrel-type distortion of PITCHf/x analytical result. Until the reliability of pitching data supplied by MLB Advanced Media is addressed, other than a fascinating visual presentation, it doesn’t really tell you much.